Prediction of Oxygen Content in Boiler Flue Gas Based on a Convolutional Neural Network

Round 1

Reviewer 1 Report

The article deals with current issues, which are topical at the moment, especially in relation to energy and mineral wealth. In the work, the authors implemented an approach using convolutional NN, on-line prediction and feaure extraction. This is also an innovation in the field. The conclusions are supported by the achieved results, in terms of the evaluated parameters R2, RMSE, etc. I have a global comment about the article: the symbols (e.g. input matrix A, 4 x 4 and many other symbols) are not aligned with the font of the text in the lines. Please fix. This is probably caused by using the equation editor. From a professional point of view, this is not an obstacle. However, from a formal point of view, I recommend this correction to the authors throughout the document. Based on the study of the article, the final assessment is: accept for acceptance, after incorporating these formal corrections.

Author Response

Reply: The English language in the revised manuscript has been carefully corrected for improved grammar and readability. The symbols (e.g. input matrix A, 4 x 4 and many other symbols) are aligned with the font of the text in the lines.

Reviewer 2 Report

This manuscript presented a proposed model based on CNN to predict flue gas oxygen content. The methodology is clear and the content is well organized. Please see my comments as follows.

1.     Does the coal composition change affect the model prediction accuracy?

2.     Is the model capable for predicting different types of boilers other than the CFB presented in the manuscript?

3.     How does the model perform in off-design conditions?

4.     How do the authors think the methodology or model described here can be migrated for different types of power systems performance prediction? Or, the model is only specific for predicting flue gas content in coal-fired power plants?

5.     Some minor items:

-       Please correct the Table 2 second column unit.

-       What are the reasons to cause the uncertainties in Fig. 7 (b)?

Thanks.

Author Response

1. Does the coal composition change affect the model prediction accuracy?

   Reply: The combustion of the boiler is typically impacted by changes in coal composition, and if the data used for modeling do not change accordingly, the accuracy of the model will be affected. In this study, a time series model (TS-CNN) for boiler flue gas oxygen content prediction is developed, and the data utilized for modeling are τ samples from the boiler's historical moment n-τ+1 to the present moment n. These data are constantly being updated, which can complement new data created by the system after the coal type is changed in a timely manner and increase the model's accuracy.

2.   Is the model capable for predicting different types of boilers other than the CFB presented in the manuscript?

   Reply: The proposed TS-CNN model in this study may theoretically be employed to predict the flue gas oxygen content of a wide range of boilers, including CFB boilers. In our future research, we'll also make an effort to confirm the performance of the proposed method for predicting the flue gas oxygen content of different types of boilers.

3. How does the model perform in off-design conditions?

   Reply: The historical operation sample data acquired in this study covers the boiler's 60%–100% load circumstances and contains the important characteristics of the boiler during operation and it can make the proposed TS-CNN model more consistent with the actual process. The performance of the model under off-design conditions depends on the specific task and the quality and quantity of the data provided. In general, model predictions under off-design conditions may not perform as accurately or reliably as under calibration conditions.

4.  How do the authors think the methodology or model described here can be migrated for different types of power systems performance prediction? Or, the model is only specific for predicting flue gas content in coal-fired power plants?

   Reply: The boiler system is a complicated nonlinear system with time lag, where the current value of each operational variable does not accurately describe the current operating state of the boiler, which may be the accumulation of states over a period of time in the past. therefore, the TS-CNN model proposed in this study takes into account the changes in the state of the boiler over a period of time, which is useful for improving the accuracy of the prediction and It is also potentially applicable to other system performance prediction of boilers.

5.  Some minor items: Please correct the Table 2 second column unit.  What are the reasons to cause the uncertainties in Fig. 7 (b)?

   Reply: We corrected the second column of the table header in Table 2. The test results of the various models used in this study are shown in Figure 7(b). It is evident that the TS-CNN model has the best fitting effect, followed by the CNN model. Both the BPNN model and the LSSVM model fit the data poorly. The following are possible reasons: 1. The boiler has the state accumulation feature; 2. The oxygen content in boiler flue gas is not only related to other variables, but also affected by its own historical change trends; 3. The feature extraction module in CNN can extract time sequence features from the input matrix.

Reviewer 3 Report

The paper presents the development of a CNN model to predict the oxygen content in boiler flue gas. The paper subject is interesting. The model is applied on a real operating boiler analysis.

However, the use of CNN to predict the performance of equipment is not a novelty. The authors must discuss the paper novelty taking in view some gaps detected based on bibliographical review.

The CNN proposed by the authors define the oxygen content in flue gas based on other operational variables. As the oxygen content is a monitored parameter, for what analysis may the model applied?

In table 1 what is the meaning of the scope column?

The authors must clarify how the network hyperparameters, presented in table 3, were defined.

The cross-validation procedure was used in the training set?

Author Response

1.The use of CNN to predict the performance of equipment is not a novelty. The authors must discuss the paper novelty taking in view some gaps detected based on bibliographical review.

Reply: The contribution of this study is to develop a time series model (TS-CNN) for boiler flue gas oxygen content prediction using samples from the historical moment n-τ+1 to the current moment n of the boiler. The TS-CNN model proposed accounts for the variations in the boiler's condition over time, which helps to increase forecast accuracy. Results from the case study demonstrate that the TS-CNN model outperforms the CNN model, LSSVM model, and BPNN model in terms of prediction accuracy.

2. The CNN proposed by the authors define the oxygen content in flue gas based on other operational variables. As the oxygen content is a monitored parameter, for what analysis may the model applied?

Reply: Boiler flue gas oxygen content can be measured using either a direct measurement method or a soft measuring approach. In order to detect oxygen content, the direct measuring method mostly uses zirconia sensors, which has drawbacks including measurement time lag, low measurement precision, high hardware replacement cost, and short equipment life. In this study, we proposed a soft measurement method based on boiler operation data. Considering that the boiler flue gas oxygen content is affected by its own historical trend, we added the historical values of boiler flue gas oxygen content to the input of the model for the prediction of boiler flue gas oxygen content in the next moment.

3. In table 1 what is the meaning of the scope column?

Reply: In this study, 12,000 historical operating samples were collected for modeling. Each sample consists of boiler flue gas oxygen content and 23 key operating variables. "scope" represents the upper and lower bounds of each variable.

4. The authors must clarify how the network hyperparameters, presented in table 3, were defined. The cross-validation procedure was used in the training set?

Reply: The hyper-parameters of the TS-CNN model include maximum number of iterations, sample size of minimum training batch, initial learning rate, learning rate decline factor, learning rate decline frequency interval, discard rate, optimization algorithm, etc. The optimal values of the hyper-parameters are determined through several experiments as well as empirical values. Cross-validation was performed throughout the test. The 80% of the data in the training set was used to create a new training set and the remaining 20% of the data was utilized as a validation set, and each iteration accompanied by one validation of the model parameters.

Round 2

Reviewer 3 Report

The authors revised the paper according to the reviewer comments.

The paper technical explanations were improvaed.