LSTM-Based Model-Predictive Control with Rationality Verification for Bioreactors in Wastewater Treatment

Round 1

Reviewer 1 Report

The authors studied the LSTM-based model-predictive control with safety verification (LSTM-MPC) for the real-time control of Anaerobic-Anoxic-Oxic process to optimize the control of aeration volume, internal recirculation, and sludge internal recycle processes for both saving energy and maintaining the stability of the bioreactor operation. These methods are validated through data from a real-world Urban Wastewater Treatment Plants in eastern China.

Comments:

1. What is written in line 115 & 116 (in Chinese). Correct it.

2.         2. Please check the language of the paper carefully.   

3.         3. Give the limitations of the study in detail.

4.         4. What are the advantages of the proposed LSTM-MPC model.

5.         5. Figure 6 & 7 are not clear. Redraw them clearly.

6.         6. Conclusion should be concise and brief.

Please check the language of the paper carefully.

Author Response

Dear Editor, Dear Reviewers,

Thanks for your review. We consider the comments and provide a new version of the manuscript. Please find our point-by-point reply below.

Yours sincerely,

Kunlun Xin, Yuting Liu

College of Environmental Science and Engineering, Tongji University.

Author Response File: Author Response.docx

Reviewer 2 Report

The study develops a machine-learning-based model-predictive control (LSTM-MPC) with rationality verification for the real-time control of the Anaerobic-Anoxic-Oxic (AAO) process in a Wastewater Treatment Plant. The authors found that LSTM-MPC with rationality verification can reduce oxygen consumption by 7% and achieve a stable control trajectory simultaneously. The manuscript is organized well and easy to read. I recommend the publication of the manuscript if the authors address the following minor comments.

Minor comments:

Lines 2-3: Too many acronyms in the title!

Line 14: Define LSTM at its first appearance

Line 22: Define MSE and NSE

Line 37: How about the energy consumption of other countries? It has to be much lower than 0.29 kWh/m3 to conclude that China has substantial potential for energy conservation.

Line 48: Define ASM1

Lines 66-68: Aren't ANN and DNN the same technique? ANN is a type of machine learning.

Line 90: What do you mean by extant studies?

Figure 1: "efluent" to "effluent"

Lines 115-116: What do you mean here?

Equation 4: Delete Xt+deltat?

Lines 172-174: Move them to the caption of Figure 2.

Lines 289-291: What did you do if more than one d(x,y) is less than the threshold?

Section 4.1: I didn't find your test performance but only the training performance for LSTM in Table 4 and Figure 6.

Line 339: What do you mean by saying "the process simulation is large"?

Line 376: "in 24 48 hours"?

Line 384: Figure 8 shows an increase in DO1 after LSTM-MPC.

Line 388: "imprve" to "improve"

Line 400: How much was the overall energy consumption reduced?

Line 401-402: Did you refer to LSTM-MPC or LSTM-MPC with rationality verification?

Line 530-533: 7% is not a significant improvement.

Author Response

Dear Editor, Dear Reviewers,

Thanks for your review. We consider the comments and provide a new version of the manuscript. Please find our point-by-point reply below.

Yours sincerely,

Kunlun Xin, Yuting Liu

College of Environmental Science and Engineering, Tongji University.

Author Response File: Author Response.docx