Cogeneration load forecasting and optimization control method

Abstract

The application discloses a method for cogeneration load prediction and optimization control and relates to the technical field of cogeneration optimization control. The importance of each variable in the original data set is obtained through different methods, and dimension reduction of the characteristic variables can better select important features to improve the model performance and then improve the universality and reduce the overfitting risk. By comparing the performance of different time series models to select the optimal model, the production load of the thermal power plant can be predicted with high accuracy. Combined with coal quantity change calculation and the time delay of the influence of coal quantity change on steam quantity, the production steam pressure can be effectively stabilized, the production demand can be met, the equipment loss can be reduced, and the main problem long plagued by the thermal power plant can be permanently solved.

Description

Technical Field

[0001] This invention relates to the field of combined heat and power (CHP) optimization control technology, specifically to a method for CHP load prediction and optimization control. Background Technology

[0002] Combined heat and power (CHP) is a comprehensive energy utilization method that combines heating and power generation, utilizing the latent heat of vaporization for both production and domestic heating while generating electricity. Due to its significant economic and environmental benefits, CHP has experienced rapid development and a construction boom in my country, driven by strong policy support. However, in actual production processes utilizing latent heat of vaporization, frequent fluctuations in steam consumption or intermittent production lead to frequent changes in steam supply pressure, which is detrimental to industrial production. Furthermore, the boiler side experiences wear and tear as it tries to meet the increased steam supply pressure. With the development of the CHP industry, stabilizing steam supply pressure has become a major challenge. Currently, operators rely primarily on experience for coordination and control, which has significant limitations and is not very effective. To ensure that the steam supply pressure meets the needs of production and daily life, it is necessary to combine the characteristics of steam consumption in the power plant and predict the changes in the amount of steam required for production and daily life at the next moment by forecasting the production load. The boiler side can then adjust the amount of coal consumed in advance based on the changes in the amount of steam supplied. This optimized control can achieve the purpose of stabilizing the steam pressure for production and daily life and protecting the equipment. Summary of the Invention

[0003] The purpose of this invention is to predict the change in the amount of steam required for production at the next moment by forecasting the production load, and to adjust the amount of coal in advance by adjusting the amount of steam on the boiler side according to the change in the amount of steam, so as to provide a method for load prediction and optimization control of cogeneration, thereby achieving stable regulation of steam supply pressure in cogeneration plants and reducing equipment wear.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: a method for load prediction and optimization control of combined heat and power (CHP), characterized by comprising the following steps:

[0005] S1. Obtain turbine operating data, preprocess the data, including data cleaning and data transformation, to obtain the original dataset;

[0006] S2. Analyze the importance of each variable in the original dataset, delete variables that have no significant impact on the total steam volume and their corresponding data, and obtain the important features;

[0007] S3. Divide the data processed in step S2 into a training dataset and a test dataset; construct a time series model, input the training dataset into the time series model for training, and obtain the trained time series model.

[0008] S4. Input the test dataset into the time series model trained in step S3 to obtain the predicted data and test the model's performance.

[0009] S5. Predict the required steam volume of 1.6MPa and 4.3MPa at the next moment, and add the prediction results to obtain the predicted total steam volume.

[0010] S6. Obtain the operating data of the boiler steam generation process, preprocess the data, fill in missing data, filter out abnormal data, and obtain the original dataset.

[0011] S7. Analyze the importance of each variable in the original dataset, delete variables and their corresponding data that have no significant impact on the steam output of the boiler, and obtain the important features.

[0012] S8. Divide the data processed in step S7 into a training dataset and a test dataset; construct a support vector regression model, input the training dataset into the model for training, and obtain the trained model.

[0013] S9. Input the test dataset into the model trained in step S8 to obtain the prediction data and test the model's performance.

[0014] S10. Obtain the coal feed rate by using the total steam volume in step S5 through a support vector regression model.

[0015] S11. Analyze the boiler data to obtain the time delay of the effect of the change in coal feed on the steam output; based on the time delay, increase the amount of coal obtained in step S10 in advance.

[0016] A further technical solution involves obtaining the steam turbine and production-end steam data in step S1, the specific process of which is as follows:

[0017] S1-1. Fill in missing data and filter out abnormal data;

[0018] S1-2. Convert the timestamps from every 5 seconds to every 30 seconds, 1 minute, 5 minutes, and 10 minutes. Visualize the total steam consumption of 1.6MPa and 4.3MPa using different timestamps to analyze the steam consumption patterns of 1.6MPa and 4.3MPa, thereby increasing the prediction accuracy in subsequent steps.

[0019] A further technical solution is to analyze the importance of each variable in the original dataset in step S2, use the Pearson correlation coefficient method in the filtering method to select features, retain the data with strong correlation, delete the data with weak correlation, and obtain a subset of the original features;

[0020] Using the Pearson coefficient as the feature scoring standard, the larger the absolute value of the correlation coefficient, the stronger the correlation. That is, the closer the correlation coefficient is to 1 or -1, the stronger the correlation; the closer the correlation coefficient is to 0, the weaker the correlation. The specific formula is:

[0021]

[0022] Where X represents the operating data of all steam turbines, Y represents the total steam volume, E represents the expected value, and μ represents the average value.

[0023] The final results show nine characteristics: steam turbine to circulating water pump flow rate, steam turbine distributor cylinder external grid steam supply flow rate (4.3 MPa), steam turbine to No. 1 high-pressure heater inlet steam flow rate, steam turbine to No. 2 high-pressure heater inlet steam flow rate, steam turbine inlet steam flow rate, No. 1 desuperheater and pressure reducer outlet and total extraction steam flow rate (1.6 MPa), No. 1 desuperheater and pressure reducer inlet flow rate, No. 2 desuperheater and pressure reducer inlet flow rate, and steam turbine to asynchronous generator flow rate.

[0024] A further technical solution is to construct a time series model in step S3 by training the training dataset through an LSTM time series model with a double hidden layer neural network structure, wherein the first layer has 500 neurons and the second layer has 100 neurons, thereby training the LSTM time series model.

[0025] A further technical solution is to detect the accuracy of the time series model in step S4, using the root mean square error (RMSE) and the goodness of fit (R²). 2 The accuracy of the model is measured using the following formula:

[0026] Root Mean Square Error (RMSE):

[0027]

[0028] in, The value is the difference between the actual value and the predicted value on the test set. The smaller the value, the better the model performance.

[0029] Goodness of fit R 2 :

[0030]

[0031] Where the numerator represents the sum of squares of the differences between the actual and predicted values, and the denominator represents the sum of squares of the differences between the actual and the mean values, according to R... 2 The value range is [0, 1], and the larger the value, the better the model.

[0032] A further technical solution is to analyze the importance of each variable in the original dataset in step S7, use the Pearson correlation coefficient method in the filtering method to select features, retain data with strong correlation and delete data with weak correlation, and obtain a subset of the original features by the same calculation method as in step S2.

[0033] A further technical solution is to construct a support vector regression model in step S8, with the original form of the objective function as follows:

[0034]

[0035] Where m is the number of samples, our samples are (x1, y1), (x2, y2), ..., (x m ,y m w, b are the regression hyperplanes wx i +b = 0 coefficient, Let φ(x) be the relaxation coefficient for the i-th sample, C be the penalty coefficient, and ε be the loss boundary. Points in the training set whose distance to the hyperplane is less than ε have no loss. i ) is a mapping function from low dimension to high dimension.

[0036] The Lagrangian function and the dualized form are as follows:

[0037]

[0038] Among them, those different from the original form Let K(x) be the Lagrange coefficient vector. i ,x j ) represents the kernel function to be used.

[0039] A further technical solution is to improve the accuracy of the support vector regression model in step S9 by using the mean squared error (MSE) and the goodness of fit (R²). 2 The accuracy of the model is measured using the following formula:

[0040] Mean Square Error (MSE):

[0041]

[0042] in, The value is the difference between the actual value and the predicted value on the test set. The smaller the value, the better the model performance.

[0043] Goodness of fit R 2 :

[0044]

[0045] Where the numerator represents the sum of squares of the differences between the actual and predicted values, and the denominator represents the sum of squares of the differences between the actual and the mean values, according to R...2 The value range is [0:1], and the larger the value, the better the model.

[0046] A further technical solution involves analyzing the boiler's steam generation process data as described in step S11 to obtain the time delay of the impact of coal feed rate changes on steam output. This involves extracting 30,000 time intervals with only 1-5 coal feeding operations, and then using data visualization analysis methods to determine the time delay of coal feed rate changes on steam output in each time interval. The average of the time delays across all extracted time intervals is then taken as the time delay of boiler coal feed rate changes on steam output.

[0047] Compared with existing technologies, the beneficial effects of this invention are as follows: by obtaining the importance of each variable in the original dataset through different methods, dimensionality reduction of the feature variables can better select important features, improve model performance, and thus improve generality and reduce the risk of overfitting. By comparing the performance of different time series models and selecting the optimal model, the prediction of the production load of thermal power plants has high accuracy. Combined with the calculation of coal quantity changes and the time delay of the impact of coal quantity changes on steam quantity, the production steam supply pressure can be effectively stabilized, meeting production needs while reducing equipment wear and tear, and permanently solving the main problems that have long plagued thermal power plants. Attached Figure Description

[0048] Figure 1 This is a flowchart of the present invention.

[0049] Figure 2 The diagram shows the steam supply pressure curve and the observed value curve after optimization by this invention. Detailed Implementation

[0050] 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.

[0051] like Figure 1 As shown, the load forecasting and optimization control of combined heat and power (CHP) includes the following steps:

[0052] S1. Obtain the operating data of the steam turbine in the thermal power plant, fill in the missing data, filter out abnormal data, and convert the timestamps from every 5 seconds to 30 seconds, 1 minute, 5 minutes and 10 minutes respectively. Visualize the amount of steam used with different timestamps to analyze the pattern of steam use and increase the prediction accuracy in subsequent steps.

[0053] S2. By using the Pearson correlation coefficient method in the filtering method to select features, we retain the data with strong correlation and delete the data with weak correlation to obtain a subset of the original features.

[0054] Using the Pearson coefficient as the feature scoring standard, the larger the absolute value of the correlation coefficient, the stronger the correlation. That is, the closer the correlation coefficient is to 1 or -1, the stronger the correlation; the closer the correlation coefficient is to 0, the weaker the correlation. The specific formula is:

[0055]

[0056] Where X represents the operating data of all steam turbines, Y represents the total steam volume, E represents the expected value, and μ represents the average value.

[0057] Ultimately, nine characteristics were obtained: steam turbine to circulating water pump flow rate, steam turbine distributor cylinder external network steam supply flow rate, steam turbine to No. 1 high-pressure heater inlet steam flow rate, steam turbine to No. 2 high-pressure heater inlet steam flow rate, steam turbine inlet steam flow rate, No. 1 desuperheater and pressure reducer outlet and total extraction steam flow rate, No. 1 desuperheater and pressure reducer inlet flow rate, No. 2 desuperheater and pressure reducer inlet flow rate, and steam turbine to asynchronous generator flow rate.

[0058] S3. Divide the data processed in step S2 into a training dataset and a test dataset in a 4:1 ratio. Use an LSTM time series model with a double-hidden-layer neural network structure, where the first layer has 500 neurons and the second layer has 100 neurons. A random seed is set during training. Table 1 lists the structure used in the model, the learning rate, the activation function of the first hidden layer, the activation function of the second hidden layer, the output activation function, and the validation error.

[0059] Table 1

[0060] LSTM time series model structure Learning rate First hidden layer activated Second hidden layer activated Output activation Verification error LSTM9-400-80-1 0.01 Tanh Tanh sigmoid 3.876 LSTM9-500-80-1 0.001 ReLU ReLU sigmoid 1.542 LSTM9-400-100-1 0.001 ReLU Tanh sigmoid 0.797 LSTM9-500-100-1 0.001 Tanh ReLU sigmoid 0.081

[0061] S4. Input the test dataset into the time series model in S3 to obtain the steam supply flow rate of the turbine distributor cylinder to the external network (4.3MPa), the outlet flow rate of the No. 1 desuperheater and the total extraction steam flow rate (1.6MPa), and use the root mean square error (RMSE) and goodness of fit (R²) to calculate the results. 2 The accuracy of the model is measured using the following formula:

[0062] Root Mean Square Error (RMSE):

[0063]

[0064] in, The value is the difference between the actual value and the predicted value on the test set. The smaller the value, the better the model performance.

[0065] Goodness of fit R 2 :

[0066]

[0067] Where the numerator represents the sum of squares of the differences between the actual and predicted values, and the denominator represents the sum of squares of the differences between the actual and the mean values, according to R... 2 The value range is [0, 1], and the larger the value, the better the model.

[0068] S5. Predict the required steam volume of 1.6MPa and 4.3MPa at the next moment, and add the prediction results to obtain the predicted total steam volume.

[0069] S6. Obtain the operating data of the boiler steam generation process, preprocess the data, fill in missing data, filter out abnormal data, and obtain the original dataset.

[0070] S7. Use the Pearson correlation coefficient method in the filtering method to select features, retain the data with strong correlation and delete the data with weak correlation, and calculate the subset of the original features using the same formula as in step S2.

[0071] S8. Divide the data processed in step S7 into a training dataset and a test dataset in a 4:1 ratio. Construct a support vector regression model; the original form of the objective function is as follows:

[0072]

[0073] Where m is the number of samples, our samples are (x1, y1), (x2, y2), ..., (x m ,y m w, b are the regression hyperplanes wx i +b = 0 coefficient, Let φ(x) be the relaxation coefficient for the i-th sample, C be the penalty coefficient, and ε be the loss boundary. Points in the training set whose distance to the hyperplane is less than ε have no loss. i ) is a mapping function from low dimension to high dimension.

[0074] The Lagrangian function and the dualized form are as follows:

[0075]

[0076] Among them, the form that is different from the original form Let K(x) be the Lagrange coefficient vector. i ,x j ) represents the kernel function to be used. Table 2 lists the corresponding model performance obtained from the parameters used in the support vector regression model.

[0077] Table 2

[0078] Penalty coefficient C Kernel function Verification error <![CDATA[10 -5 ]]> poly 78.637 <![CDATA[10 -5 ]]> rbf 32.458 <![CDATA[10 -6 ]]> rbf 7.119 <![CDATA[10 -7 ]]> rbf 0.874

[0079] S9. Input the test dataset into the support vector regression model constructed in step S8, and use the mean squared error (MSE) and goodness of fit (R²) to perform the analysis. 2 This is used to measure the accuracy of the model. The specific calculation formula is as follows:

[0080] Mean Square Error (MSE):

[0081]

[0082] in, The value is the difference between the actual value and the predicted value on the test set. The smaller the value, the better the model performance.

[0083] Goodness of fit R 2 :

[0084]

[0085] Where the numerator represents the sum of squares of the differences between the actual and predicted values, and the denominator represents the sum of squares of the differences between the actual and the mean values, according to R... 2 The value range is [0, 1], and the larger the value, the better the model.

[0086] S10. The total amount of steam in step S5 is used to obtain the coal feed amount through the support vector regression model. 1.1t of coal needs to be added.

[0087] S11. Extract 30,000 time intervals with only 1-5 coal addition operations, and then use data visualization analysis methods to determine the time delay of coal quantity changes on steam quantity in each time interval. Take the average of the time delays of all extracted time intervals as 1 min 36 s, representing the time delay of boiler coal quantity changes on steam. Based on this time delay, increase the coal quantity by 1.1 t 1 min 36 s in advance, thereby achieving the goal of stable steam supply pressure to meet production and living needs.

[0088] The above data selected three months of turbine operation data and boiler steam generation process operation data from a chemical plant. The results, obtained through LSTM time series and support vector regression models, were compared with steam pressure data obtained by technicians based on experience. Figure 2 As shown, through model control and advance control, the pressure can be stabilized between 1.48MPa and 1.67MPa, achieving stable steam supply pressure to meet the needs of production and daily life.

[0089] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art can make changes using different methods or approaches. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for cogeneration load forecasting and optimal control, characterized in that The method comprises the following steps: S1, obtaining steam turbine operation data, preprocessing the data, the preprocessing comprising data cleaning and data transformation, and obtaining an original data set; S2, analyzing the importance of each variable in the original data set, deleting variables and corresponding data that have no obvious influence on the total steam quantity, and obtaining important features; S3, dividing the data processed in step S2 into a training data set and a test data set; constructing a time series model, inputting the training set into the time series model for training, and obtaining a trained time series model; S4, inputting the test data set into the trained time series model in step S3 to obtain predicted data and detect the performance of the model; S5, predicting the steam quantities required at 1.6 MPa and 4.3 MPa at the next moment, adding the prediction results to obtain the predicted total steam quantity; S6, obtaining operation data of a boiler steam generation process, preprocessing the data, filling in missing data, and screening out abnormal data to obtain an original data set; S7, analyzing the importance of each variable in the original data set, deleting variables and corresponding data that have no obvious influence on the steam quantity generated by the boiler, and obtaining important features; S8, dividing the data processed in step S7 into a training data set and a test data set; constructing a support vector regression model, inputting the training data set into the model for training, and obtaining a trained model; S9, inputting the test data set into the trained model in step S8 to obtain predicted data and detect the performance of the model; S10, obtaining the coal supply quantity through the support vector regression model based on the total steam quantity in step S5; S11, analyzing the boiler data to obtain the time delay of the influence of the coal supply quantity change on the steam quantity; and increasing the coal quantity obtained in step S10 in advance according to the time delay; The specific process of step S1 is as follows: S1-1, filling in missing data and screening out abnormal data; S1-2, converting the time stamp from every 5 seconds to 30 seconds, 1 minute, 5 minutes and 10 minutes respectively, and using different time stamps to visualize the data of the total steam quantity at 1.6 MPa and 4.3 MPa to analyze the steam consumption law of 1.6 MPa and 4.3 MPa, thereby increasing the prediction accuracy in the subsequent steps; In step S2, the importance of each variable in the original data set is analyzed, the Pearson correlation coefficient method in the filtering method is used for feature selection, the data with strong correlation is retained, and the data with weak correlation is deleted to obtain a subset of original features; The Pearson coefficient is used as a feature score standard, the larger the absolute value of the correlation coefficient, the stronger the correlation, that is, the closer the correlation coefficient is to 1 or -1, the stronger the correlation, and the closer the correlation coefficient is to 0, the weaker the correlation; the specific formula is: Wherein, X is all steam turbine operation data, Y is the total steam quantity, E is the expectation, and μ is the average; ； Finally, 9 features are obtained, including the steam turbine to circulating water pump flow, the steam turbine to external network steam supply flow of the steam turbine cylinder, the steam turbine to 1# high pressure heater steam flow, the steam turbine to 2# high pressure heater steam flow, the steam turbine inlet steam flow, the 1# temperature and pressure reducer outlet and steam extraction total flow, the 1# temperature and pressure reducer inlet flow, the 2# temperature and pressure reducer inlet flow, and the steam turbine to asynchronous generator flow. ​ The step S11 analyzes the operation data of the boiler steam generation process to obtain the time delay of the coal supply change on the steam volume, intercepts 30,000 time interval data with only 1-5 coal supply operations, and then uses a data visualization analysis method to determine the time delay of the coal supply change on the steam volume in each time interval. The average of the time delays of all the time intervals is taken as the time delay of the coal supply change on the steam generation.

2. The method for cogeneration load forecasting and optimization control according to claim 1, characterized in that: The step S3 constructs a time series model. The training data set is input into an LSTM time series model with a double-hidden layer neural network structure, where the number of first-layer neurons is 500, and the number of second-layer neurons is 100. Thus, the LSTM time series model is trained.

3. The method for cogeneration load forecasting and optimization control according to claim 1, wherein: The accuracy of the time series model detected in the step S4 is measured by the root mean square error RMSE and the goodness of fit , specifically as follows: Root mean square error RMSE: ； wherein, For true value-predicted value on the test set, the smaller the value, the better the model performance; Goodness of fit R 2 : ； where the numerator part represents the sum of squares of the difference between the true value and the predicted value, and the denominator part represents the sum of squares of the difference between the true value and the average value, according to R 2 The value range of R is [0, 1], and the larger the value is, the better the model is.

4. The method for cogeneration load forecasting and optimization control according to claim 1, wherein: In the step S7, the importance of each variable in the original data set is analyzed, the Pearson correlation coefficient method in the filtering method is used for feature selection, the data with strong correlation is retained, the data with weak correlation is deleted, and the subset of the original features is obtained according to the calculation formula in the step S2.

5. The method for cogeneration load forecasting and optimization control according to claim 1, wherein: In the step S8, a support vector regression model is constructed. The original form of the specific objective function is as follows: ； ； ； where m is the number of samples, the samples are (x1, y1), (x2, y2), …, (x m ,y m ); w, b are the coefficients of the regression hyperplane wx i + b = 0, , is the relaxation coefficient of the ith sample, C is the penalty coefficient, and ε is the loss boundary. The points of the training set with a distance to the hyperplane less than ε have no loss, is the mapping function from low dimension to high dimension; Through the Lagrange function and the dual form, the following is obtained: ； ； ； ； where the original form is different , is the Lagrange coefficient vector, is the kernel function to be used.

6. The method for cogeneration load forecasting and optimization control according to claim 1, wherein: The accuracy of the support vector regression model in the step S9 is measured by the mean square error MSE and the goodness of fit R 2 to measure the accuracy of the model.

Images