A data-driven-based power plant boiler flue gas oxygen content prediction optimization method

Abstract

The application discloses a power plant boiler flue gas oxygen content prediction optimization method based on data driving, and relates to the technical field of power plant boiler optimal flue gas oxygen content prediction optimization. For a boiler of a power plant in South China, historical data of a boiler furnace pressure, a main steam pressure, a main steam flow, an AGC load instruction, a steam drum pressure, a coal supply amount, a pulverized coal concentration, an induced draft amount, a supply air amount, an optimal flue gas oxygen content, an exhaust gas temperature, an outlet temperature of a superheater, a steam drum water level and an outlet pressure of the superheater are acquired, the historical data are preprocessed to obtain a modeling data set, a furnace pressure model, a main steam pressure model and an optimal flue gas oxygen content model are built through a support vector machine algorithm, and then controllable variables are optimized through an artificial fish swarm algorithm. The application can reduce pulverized coal consumption and save economic cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power plant boiler flue gas oxygen content prediction optimization, and particularly relates to a power plant boiler flue gas oxygen content prediction optimization method based on data driving. BACKGROUND

[0002] The AGC automatic generation control of the thermal power plant unit is essentially a high-level control system for connecting the power grid energy management system and the generator unit control system, which coordinates the operation of the boiler and the steam turbine, and is the link between the power grid dispatching and the unit and boiler control systems, and is required to ensure the rapid response to the load change of the power grid and the stability of the unit itself, which will inevitably consume a large amount of coal fuel. In order to achieve the purpose of responding to the AGC instruction while saving energy and reducing consumption, the optimal flue gas oxygen content has a great influence on the efficiency of the coal combustion in the boiler during the operation of the thermal power plant unit, and further influences the consumption of the coal, so that the optimal flue gas oxygen content is predicted, the value of the air supply is adjusted, the coal is fully combusted, the consumption of the coal is reduced, and the economic cost is saved. SUMMARY

[0003] The present application aims to provide a power plant boiler flue gas oxygen content prediction optimization method based on data driving, which fully combusts the coal by controlling the air supply, and reduces the consumption of the coal.

[0004] To solve the above technical problems, the present application adopts the following technical scheme: a power plant boiler flue gas oxygen content prediction optimization method based on data driving, characterized by comprising the following steps:

[0005] S1, obtaining the historical data of the furnace pressure, the main steam pressure, the main steam flow, the AGC load instruction, the steam drum pressure, the coal supply amount, the coal concentration, the induced draft, the air supply, the optimal flue gas oxygen content, the exhaust gas temperature, the superheater outlet temperature, the steam drum water level and the superheater outlet pressure of the power plant boiler to be predicted and optimized;

[0006] S2, performing a preprocessing operation on the historical data, including missing value, abnormal value, normalization, feature importance analysis, to obtain a modeling data set;

[0007] S3, inputting the modeling data set into a support vector machine algorithm to obtain the furnace pressure model, the main steam pressure model and the optimal flue gas oxygen content model of the power plant boiler through training, and evaluating the accuracy of the prediction model by using the mean absolute error MAE and the root mean square error RMSE;

[0008] S4, taking the furnace pressure and the main steam pressure as constraint variables, the air supply as a controllable variable, and the optimal flue gas oxygen content as a target variable, wherein the furnace pressure is in a range of -80-100 Pa, the main steam pressure is in a range of 13-17.29 MPa, and the air supply is in a range of 430-1272 t / h; the air supply is optimized by using the artificial fish swarm algorithm, so that the optimal flue gas content is achieved, the pulverized coal is fully combusted, and the consumption of the pulverized coal is reduced.

[0009] A further technical solution is the data acquisition and storage of the step S1, and the specific process is as follows:

[0010] S1-1, collecting and summarizing the historical data of a 3# boiler of a power plant in South China according to a time stamp;

[0011] S1-2, storing the data in a CSV file in a way that the self-increment type is used as a primary key, as a historical data set;

[0012] A further technical solution is the data preprocessing step of the step S2, and the specific process is as follows:

[0013] S2-1, checking whether there is missing data and abnormal value data in the historical data set by using a table tool, wherein the abnormal value data refers to data that exceeds the value range of the current column, and the missing data and the abnormal value data are supplemented and replaced by the mean value of the current column;

[0014] S2-2, each column of data in the historical data set needs to be preprocessed by using a maximum-minimum normalization method due to different dimensions, and the specific formula is as follows:

[0015]

[0016] wherein x* is the normalized value of the sample data, x is the sample data, x max is the maximum value of the sample data, and x min is the minimum value of the sample data;

[0017] S2-3, performing feature importance analysis by using an L1 regularization method, deleting the data of the drum pressure, the flue gas temperature, the superheater outlet temperature, the drum water level, and the superheater outlet pressure, and obtaining 9 variables of the furnace pressure, the main steam pressure, the main steam flow, the AGC load instruction, the coal supply, the pulverized coal concentration, the induced air volume, the air supply, and the optimal flue gas oxygen content and corresponding data as a modeling data set.

[0018] A further technical solution is the data set division, the model parameter, and the model construction of the step S3, and the specific process is as follows:

[0019] S3-1, 17280 pieces of data in the modeling data set are divided into a training set, a validation set and a test set according to a ratio of 8:1:1, wherein the training set data are 13824 pieces, the validation set data are 1728 pieces, and the test set data are 1728 pieces;

[0020] S3-2, the data parameters needed for building the furnace pressure model are the induced draft and the air supply, the data parameters needed for building the main steam pressure model are the main steam flow, the AGC load instruction, the coal supply, the flue gas oxygen content and the air supply, and the data parameters needed for building the optimal flue gas oxygen content model are the pulverized coal concentration, the coal supply, the induced draft and the air supply;

[0021] S3-3, the initial model hyperparameters svr_type, kernel function, gamma, degree, coefO, C and eps of the support vector machine algorithm are respectively 0, a linear kernel function, 0.01, 1, 0, 5 and 0.001.

[0022] S3-4, the evaluation indexes are calculated according to the test set, the true value is y, the predicted value is f, m represents the data quantity, that is, 1728, and the formula of the mean absolute error MAE is specifically:

[0023]

[0024] The formula of the root mean square error RMSE is specifically:

[0025]

[0026] A further technical solution is that step S4 adopts an artificial fish swarm algorithm to solve the optimal air supply, and the specific process is as follows:

[0027] S4-1, the population size N, the initial position of each fish, the fish field of view v, the step length step, the crowding factor & and the cycle number s are randomly initialized and set;

[0028] S4-2, the fitness value of each individual of the initial fish swarm is calculated, the state of the best fish is obtained and recorded;

[0029] S4-3, each individual is evaluated, and the foraging, grouping or tail chasing behavior is selected to update itself to obtain a new fish swarm;

[0030] S4-4, all individuals are evaluated, and if an individual is better than the record, the individual is recorded;

[0031] S4-5, when the optimal solution of the recorded individual meets the error range or reaches the cycle number, the algorithm ends, otherwise, step 3 is returned for iteration.

[0032] In this step, the initial values ​​of the artificial fish swarm algorithm parameters, population size N, fish field of view v, step size step, crowding factor &, and number of iterations s, are 50, 0.5, 0.8, 0.3, and 30, respectively. The initial position of each fish is determined by the air supply volume range [430, 1272], so the initial position of each fish is randomly generated.

[0033] Compared with existing technologies, the beneficial effects of this invention are as follows: By analyzing the importance of variables through feature selection, four unimportant variables and one highly correlated variable are removed, resulting in variables that significantly influence the construction of the furnace pressure model, main steam pressure model, and optimal flue gas oxygen content model: furnace pressure, main steam pressure, main steam flow rate, AGC load command, coal feed rate, pulverized coal concentration, induced draft rate, forced draft rate, and optimal flue gas oxygen content. Among these, furnace pressure and main steam pressure are used as constraint variables, forced draft rate as a controllable variable, and optimal flue gas oxygen content as the target variable. A model is built using support vector machines, and the artificial fish swarm algorithm is used to optimize the controllable variables, achieving complete pulverized coal combustion and reducing economic costs, resulting in a coal saving of 2.1%. Attached Figure Description

[0034] Figure 1 This is a flowchart of the boiler flue gas oxygen content prediction and optimization method in this invention.

[0035] Figure 2 This is a flowchart of the artificial fish swarm algorithm in this invention.

[0036] Figure 3 To optimize the comparison curve of the results. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0038] Example 1

[0039] Figure 1 As shown Figure 1 As shown, the specific implementation steps of the method for predicting and optimizing the oxygen content in boiler flue gas are as follows:

[0040] (1) Obtain the historical data of the furnace pressure, main steam pressure, main steam flow, AGC load instruction, steam drum pressure, coal supply amount, pulverized coal concentration, induced draft, supply air, optimal flue gas oxygen content, flue gas temperature, superheater outlet temperature, steam drum water level, and superheater outlet pressure of a certain power plant 3# boiler in South China to be predicted and optimized. The data collection time is from May 1, 2021 to October 31, 2021, and the modeling data set is generated after missing value, abnormal value, normalization, and feature importance analysis. The total number of modeling data set is 17280, of which 13824 is training set data, the time of training set data is from May 1, 2021 to September 25, 2021, which is used to build a training model; 1728 is the validation set data, the time of validation set data is from September 26, 2021 to October 13, 2021, which is used to adjust the training model; 1728 is the test set data. The time of test set data is from October 13, 2021 to October 31, 2021, which is used to test the prediction model;

[0041] (2) The historical data is preprocessed, including missing value, abnormal value, normalization, and feature importance analysis, to obtain the modeling data set;

[0042] (3) The modeling data set is divided into training set, validation set and test set and input into the support vector machine algorithm, and the furnace pressure model, main steam pressure model and optimal flue gas oxygen content model of the power plant boiler are obtained by training. The mean absolute error MAE and root mean square error RMSE are used to evaluate the accuracy of the prediction model;

[0043] (4) The furnace pressure and main steam pressure are taken as constraint variables, the supply air is taken as controllable variable, and the optimal flue gas oxygen content is taken as target variable, wherein the furnace pressure value range is -80~100Pa, the main steam pressure value range is 13~17.29MPa, and the supply air value range is 430~1272t / h. The artificial fish swarm algorithm is used to optimize the feature data supply air to achieve the optimal flue gas content, so that the pulverized coal is fully burned and the consumption of pulverized coal is reduced.

[0044] A further technical solution is the data acquisition and storage of step (1), and the specific process is as follows:

[0045] (1.1) Collect and summarize the historical data of a certain power plant 3# boiler in South China according to the time stamp;

[0046] (1.2) Store the data in the CSV file as the primary key in the form of self-increment, as the historical data set;

[0047] A further technical solution is the data preprocessing step of step (2), and the specific process is as follows:

[0048] (2.1), find out whether there are missing data and abnormal value data in the historical data set through the table tool, the abnormal value data refers to the value beyond the current column, the missing data and the abnormal value data are supplemented and replaced by the mean value of the current column;

[0049] (2.2), due to the different dimensions of the data values in each column in the historical data set, the maximum and minimum normalization method needs to be used for preprocessing, and the specific formula is:

[0050]

[0051] Wherein, x* is the normalized value of the sample data, x is the sample data, x max is the maximum value of the sample data, and x min is the minimum value of the sample data;

[0052] (2.3), the feature importance analysis is carried out through the L1 regularization method, the data of drum pressure, flue gas temperature, superheater outlet temperature, drum water level and superheater outlet pressure are deleted, and 9 variables of furnace pressure, main steam pressure, main steam flow, AGC load instruction, coal supply amount, pulverized coal concentration, induced draft, air supply and optimal flue gas oxygen content and corresponding data are obtained as modeling data set.

[0053] Further technical solutions are the data set division, model parameter and model construction of the step (3), and the specific process is as follows:

[0054] (3.1), 17280 data in the modeling data set are divided into a training set, a validation set and a test set according to the ratio of 8:1:1, wherein the training set data is 13824, the validation set data is 1728, and the test set data is 1728;

[0055] (3.2), the data parameters needed for the establishment of the furnace pressure model are induced draft and air supply, and the values of the final model hyperparameters svr_type, kernel function, gamma, degree, coefO, C and eps obtained through the support vector machine algorithm are 0, linear kernel function, 0.01, 3, 0, 1 and 0.01 respectively.

[0056] (3.3), the data parameters needed for the establishment of the main steam pressure model are main steam flow, AGC load instruction, coal supply amount, flue gas oxygen content and air supply, and the values of the final model hyperparameters svr_type, kernel function, gamma, degree, coefO, C and eps obtained through the support vector machine algorithm are 0, linear kernel function, 0.01, 3, 0, 10 and 0.001 respectively.

[0057] (3.4) The data parameters needed to build the optimal flue gas oxygen content model are the pulverized coal concentration, coal supply, induced draft and air supply. The support vector machine algorithm is used to obtain the final model super parameter svr_type, kernel function, gamma, degree, coefO, C and eps, whose values are 0, linear kernel function, 0.001, 3, 0, 5, 0.05 respectively.

[0058] (3.5) According to the test set, the evaluation index is calculated, the true value is y, the predicted value is f, and m represents the data amount, that is, 1728. The formula of the mean absolute error MAE is as follows:

[0059]

[0060] The formula of the root mean square error RMSE is as follows:

[0061]

[0062] The evaluation index values of the furnace pressure model MAE and RMSE are 0.0125 and 1.0136 respectively, the evaluation index values of the main steam pressure model MAE and RMSE are 0.0101 and 1.0018 respectively, and the evaluation index values of the furnace pressure model MAE and RMSE are 0.0116 and 1.0007 respectively.

[0063] As shown in Figure 2 , a further technical solution is that the step (4) adopts an artificial fish swarm algorithm to solve the optimal air supply, and the specific process is as follows:

[0064] (4.1) Randomly initialize the population size N, the initial position of each fish, the fish visual field v, the step length step, the crowding factor & and the cycle number s;

[0065] (4.2) Calculate the fitness value of each individual in the initial fish swarm, obtain the state of the best fish and record it;

[0066] (4.3) Evaluate each individual, and select the foraging, grouping or tail chasing behavior to update itself to obtain a new fish swarm;

[0067] (4.4) Evaluate all individuals, and if an individual is better than the record, record the individual;

[0068] (4.5) When the optimal solution of the recorded individual meets the error range or reaches the cycle number, the algorithm ends, otherwise return to step 3 for iteration.

[0069] In this step, the final parameter population size N, fish field of view v, step, crowding factor & of the artificial fish swarm algorithm are set to 50, 0.5, 0.7, 0.3, and 20 respectively, wherein the initial position of each fish is determined by the air supply value range [430, 1272], so the initial position of each fish is randomly generated.

[0070] In order to verify the influence of the optimal oxygen content of flue gas on the full combustion of pulverized coal and reduce the consumption of pulverized coal, the data of each day from November 1, 2021 to November 6, 2021 are obtained, a total of 241 data, while keeping other operation variables unchanged, by adjusting the air supply to achieve the optimal oxygen content of flue gas, the pulverized coal is fully combusted, and the optimization result of the consumption of pulverized coal is as shown in Figure 3 The coal consumption can be saved by 2.1%.

[0071] The above-described embodiments are only a preferred scheme of the present application, and are not intended to limit the present application. Therefore, any technical solutions obtained by equivalent replacement or equivalent transformation fall within the protection scope of the present application.

Claims

1. A data-driven method for predicting and optimizing oxygen content in power plant boiler flue gas, characterized in that... Includes the following steps: S1. Obtain historical data on the boiler furnace pressure, main steam pressure, main steam flow, AGC load command, drum pressure, coal feed rate, pulverized coal concentration, induced draft rate, forced draft rate, optimal flue gas oxygen content, exhaust gas temperature, superheater outlet temperature, drum water level, and superheater outlet pressure of the power plant to be predicted and optimized. S2. Perform preprocessing operations on historical data, including missing values, outliers, normalization, and feature importance analysis, to obtain the modeling dataset; S3. Divide the modeling dataset into training, validation and test sets and input them into the support vector machine algorithm. Through training, obtain the furnace pressure model, main steam pressure model and optimal flue gas oxygen content model of the power plant boiler. Use the mean absolute error (MAE) and root mean square error (RMSE) to evaluate the accuracy of the prediction model. S4. Using furnace pressure and main steam pressure as constraint variables, air supply volume as a controllable variable, and optimal flue gas oxygen content as the target variable, the furnace pressure ranges from -80 to 100 Pa, the main steam pressure ranges from 13 to 17.29 MPa, and the air supply volume ranges from 430 to 1272 t / h. The air supply volume of the characteristic data is optimized using the artificial fish swarm algorithm to achieve the optimal flue gas content, ensuring complete combustion of pulverized coal and reducing pulverized coal consumption. The specific process of step S1 is as follows: S1-1. Collect and summarize historical data of boiler #3 of a power plant in South China according to timestamps; S1-2. Store the data in a CSV file with an auto-incrementing primary key as a historical dataset; The specific process of step S2 is as follows: S2-1. Use table tools to check for missing data and outlier data in the historical dataset, delete outlier data that exceeds the value range of the current column, and supplement and replace missing data and outlier data with the mean of the current column. S2-2. Due to the different units of measurement, the data values ​​in each column of the historical dataset are preprocessed using the min-max normalization method. The specific formula is as follows: ; Where x* is the normalized value of the sample data, x is the sample data, and x is the normalized value of the sample data. max x is the maximum value of the sample data. min The minimum value of the sample data; S2-3. Feature importance analysis was performed using the L1 regularization method. Data on drum pressure, flue gas temperature, superheater outlet temperature, drum water level, and superheater outlet pressure were deleted. Nine variables and their corresponding data were obtained as the modeling dataset for building the prediction model: furnace pressure, main steam pressure, main steam flow, AGC load command, coal feed rate, pulverized coal concentration, induced draft rate, forced draft rate, and optimal flue gas oxygen content. The specific process of step S3 is as follows. S3-1. Divide the 17280 data points in the modeling dataset into a training set, a validation set, and a test set in a ratio of 8:1:

1. The training set contains 13824 data points, the validation set contains 1728 data points, and the test set contains 1728 data points. S3-2. The data parameters used to build the furnace pressure model are induced draft and forced draft. The data parameters used to build the main steam pressure model are main steam flow, AGC load command, coal feed rate, flue gas oxygen content, and forced draft. The data parameters used to build the optimal flue gas oxygen content model are pulverized coal concentration, coal feed rate, induced draft and forced draft. S3-3. Prediction is performed using the support vector machine algorithm. The initial model hyperparameters svr_type, kernel function, gamma, degree, coefO, C, and eps are 0, linear kernel function, 0.01, 1, 0, 5, and 0.001, respectively. S3-4. Based on the test set, make predictions and calculate the evaluation index, where the true value is y, the predicted value is f, and m represents the data volume, i.e., 1728. The specific formula for the Mean Absolute Error (MAE) is: ; The formula for the root mean square error (RMSE) is as follows: .

2. The data-driven method for predicting and optimizing oxygen content in power plant boiler flue gas according to claim 1, characterized in that: Step S4 uses the artificial fish swarm algorithm to solve for the optimal air supply volume. The specific process is as follows. S4-1. Randomly initialize the population size N, the initial position of each fish, the fish's field of vision v, the step size step, the crowding factor &, and the number of loops s. S4-2. Calculate the fitness value of each individual in the initial fish group, obtain the optimal fish state and record it; S4-3. Evaluate each individual and select them to engage in foraging, grouping, or tail-chasing behaviors to update themselves and obtain a new school of fish; S4-4. Evaluate all individuals. If an individual is better than the recorded one, then record that individual. S4-5. When the optimal solution of an individual recorded is within the error range or the number of iterations is reached, the algorithm ends; otherwise, return to step 3 for iteration. In this step, the initial settings for the artificial fish swarm algorithm parameters, population size N, fish field of view v, step size step, crowding factor &, and number of iterations s, are 50, 0.5, 0.8, 0.3, and 30, respectively. The initial position of each fish is determined by the air supply volume range [430, 1272], and the initial position of each fish is randomly generated.

Images