Indoor Emission Sources Detection by Pollutants Interaction Analysis
, Lei Song 2, Abdolhossein Sarrafzadeh 3, Guy Coulson 4, Ian Longley 4 and Gustavo Olivares 4Round 1
Reviewer 1 Report
This paper proposes a method for detecting the emission sources in an indoor environment. In experiments, the NIWA-developed device is used for collecting seven activities in an indoor situation. The proposed correlation coefficient-based emission source detection is well evaluated and analyzed.
Overall, the paper is well organized, and the results are very good. The topic is very important. All information is completed in the body of the paper. Therefore, the reviewer recommends the paper is accepted as it is.
Author Response
Answer: Thanks for your positive comments on our work presented in this paper.
Reviewer 2 Report
This article presents employs the correlation coefficients technique for spectral features extraction and for detection of emission sources based on pollutant interaction analysis. Typical feature extraction methods such as PCA and LDA are also applied. However, while reading this manuscript, I could not find any clarity. The study is lacking references, clear explanations where needed. The results and conclusion are not clear.
The study lacks an appropriate experimental description of four example how the seven scenarios were carried out. Furthermore, there is no explanation about how the experimental parameters selected (PM, CO, CO2) could provide information related to pollutant interaction analysis or their expected variations in the scenarios under study. There is no information about how these parameters were collected. They do not even provide information about PM size (was it PM2.5, PM10,..).
Some specific comments:
- How is E, expected value determined in Eq. 3
- Line 1 of Eq. 3: the square root of the Var (a,j)^2 is Var(a,j). This equation is unclear: in the second line E(ai,j-E(ai,j))^2. This concept should be explained. Third line inverts the subindexes in the numerator (aj,i)
- Line 184 f*,not defined
- Line 187, why do they mean by training set?
- Sections 2.3, 2.3.2., 2.3.3: Is the mathematical model developed by the authors? Should no reference be provided?
- Figure 3: No clear conclusions are evident from these plots. I am very surprised that Smoking did not seem to increase PM, CO, CO2 in Figure 3 c
- Table 3: No information on how accuracy is calculated
- Figure 4: No explanation of how growth is calculated
- Conclusions unclear
Some minor gramatical/spelling errors:
- Line 121: "the" repeated twice
- Line 146: "an data"
- Line 170 missing the in "we define correlation model"
Author Response
#Reviewer 2
This article presents employs the correlation coefficients technique for spectral features extraction and for detection of emission sources based on pollutant interaction analysis. Typical feature extraction methods such as PCA and LDA are also applied. However, while reading this manuscript, I could not find any clarity. The study is lacking references, clear explanations where needed. The results and conclusion are not clear.
Answer: Thanks for your comments. In the revised manuscript, we clarified the key logic our work as, Indoor Air Quality, PACMAN, Data Acquisition, Model, Results, Interpolation and Conclusion. For each topic, we have revised unclear explanation added necessary references.
The study lacks an appropriate experimental description of four example how the seven scenarios were carried out. Furthermore, there is no explanation about how the experimental parameters selected (PM, CO, CO2) could provide information related to pollutant interaction analysis or their expected variations in the scenarios under study. There is no information about how these parameters were collected. They do not even provide information about PM size (was it PM2.5, PM10,..).
Answer: Thanks for this useful comment. As the air parameter data were captured by the PACMAN device. In the revised manuscript, we have added PACMAN dust sensor explanation in Table 1, in which we have identifies the particle matter size as PM10. Please find the revision in Section 2.1
Some specific comments:
- How is E, expected value determined in Eq. 3
Answer: Thanks for the comment. In probability theory, the expected value of a random variable X, denoted E(X), is a generalization of the weighted average, and is intuitively the arithmetic mean of a large number of independent realizations of X. In the revised manuscript, E was explained as the mathematical expectation.
- Line 1 of Eq. 3: the square root of the Var (a,j)^2 is Var(a,j). This equation is unclear: in the second line E(ai,j-E(ai,j))^2. This concept should be explained. Third line inverts the subindexes in the numerator (aj,i)
Answer: Thanks for the comments. Line 1 of Eq. 3, it’s the covariance of two data streams divided by the square root of the dot product between two streams variances. Here, we have corrected one typo in the revised manuscript.
Line 2 and Line3 of Eq. 3 are two steps mathematical derivation according to the probability theory and statistics
- Line 184 f*,not defined
Answer: Thanks for this comments. f* is the optimal solution to the problem of emission event detection, which has been detailed in the following section titled “Support vector machine (SVM) classification” In other words, the optional solution is, a Gaussian RBF kernel SVM function with parameter tuned via cross validation tests.
- Line 187, why do they mean by training set?
Answer: Thanks for the question. For Support Vector Machine (SVM) classifier, it needs a training before for testing. Thus, Line 187 D is given as the training data set.
- Sections 2.3, 2.3.2., 2.3.3: Is the mathematical model developed by the authors? Should no reference be provided?
Answer: Thanks for this comments. In the revised manuscript, Section 2.3.3 we have added the reference of support vector machine (SVM).
- Figure 3: No clear conclusions are evident from these plots. I am very surprised that Smoking did not seem to increase PM, CO, CO2 in Figure 3 c
Answer: Thanks for the comments. Figure 3 visualizes the effectiveness of the proposed correlation method in extracting discriminant features for emission source detection. The left column figure (a) (c) (e) plot the original values of pollutants, and the right column figure (b) (d) (f) shows the correlation coefficients of a pair of pollutants.
Smoking did have a clear increase on PM and CO. The reason Figure 3 is showing clearly such increase, because the level of CO2 is 3500, while the level of PM and CO is in the range of 100~300, we plot PM, CO, CO2 in the same figure, PM and CO data is supressed at the bottom of the figure.
- Table 3: No information on how accuracy is calculated
Answer: Thanks for this comments. In the revised manuscript, we have added the explanation of three tabled measurements: accuracy, stdev, and growth. Please find the revision in
- Figure 4: No explanation of how growth is calculated
Answer: Thanks for this useful comments. Figure 4 gives the average accuracy and standard deviation variation under the condition of different sliding window sizes. The growth is calculated in Table 4. Thus in the revised manuscript, we removed growth as a performance measurement in Figure 4 explanation, and added it in Table 4 explanation. Please find the two revisions on Page 16 and 17, respectively.
- Conclusions unclear
Answer: Thanks for the comments. We have done a careful editing of the conclusion section.
Please find the revised conclusion on Page 18.
Some minor gramatical/spelling errors:
- Line 121: "the" repeated twice
- Line 146: "an data"
- Line 170 missing the in "we define correlation model"
Answer: Thanks for the comments. In the revised manuscript, we have corrected the above errors.
Reviewer 3 Report
Title:
Indoor Emission Sources Detection by Pollutants Interaction Analysis
The main points:
- To analyse the secondary pollution.
- To calculate intra pollutants correlation coefficients for characterizing and distinguishing emission events.
- Improvement of the detection accuracy by a support vector machine.
Point 1:
The authors do not focus on explaining the scientific meaning of their results. The discussion section is not focused on comparing the results of this study with the literature data.
Point 2:
How was estimated the relative error of simulation error over the range of the experimental conditions?
Point 3:
Is there any similar device in the world to compare the results of physical experiment? If not, how the self-designed experimental system was validated?
Point 4:
Could the authors write the uncertainty of the physical model results?
Point 5:
Is the experimental procedure used for standard or self-developed?
The paper is within the scope of the journal and have the quality required for publication in the Energies journal. The present paper is clear and consistent. Therefore, in the current form, after answering to all mentioned issues, I can recommend it for publication.
Author Response
#Reviewer 3
The main points:
- To analyse the secondary pollution.
- To calculate intra pollutants correlation coefficients for characterizing and distinguishing emission events.
- Improvement of the detection accuracy by a support vector machine.
Point 1:
The authors do not focus on explaining the scientific meaning of their results. The discussion section is not focused on comparing the results of this study with the literature data.
Answer: Thanks for the comments. The focus of the presented work is to develop a pollutant interaction data analytic method for emission source detection. Under the same condition of PACMAN data, we compare the proposed method with PCA, LDA and without data filtering.
To discover scientific meaning and compare indoor emission detection with previous work reported in literature will address the optimization of the entire system including PACMAN, data filtering and classifier, three modules.
Point 2:
How was estimated the relative error of simulation error over the range of the experimental conditions?
Answer: Thanks for the comments. We estimated errors for detecting seven types of emission events via the confusion matrix, which includes truth positive, truth negative, false positive and false negative errors for all 7 type of events. Please find the information in Table 3.
Point 3:
Is there any similar device in the world to compare the results of physical experiment? If not, how the self-designed experimental system was validated?
Answer: Thanks for the comments. PACMAN device is a multi-sensor box invented by the Air quality group, National Institute of Water and Atmosphere Research, New Zealand. So far, we are not aware of any similar systems reported in literature. The system has been validated before the experiments for this work. Please find the information about how the system been validated via https://niwa.co.nz/atmosphere/research-projects/pacman
Point 4:
Could the authors write the uncertainty of the physical model results?
Answer: Thanks for the comments. The aim of this work is to develop a data analytic model for effective indoor emission event classification. The physical model is a bigger picture of the research, which can be left as a valuable future work.
Point 5:
Is the experimental procedure used for standard or self-developed?
Answer: Thanks for the comments. The procedure used in the experiment is self-developed.
The paper is within the scope of the journal and have the quality required for publication in the Energies journal. The present paper is clear and consistent. Therefore, in the current form, after answering
Reviewer 4 Report
The suggestions and changes needed in this article are:
- all bibliographic references to be numbered and specified in the article;
- all bibliographic references don't respect the templete
Author Response
Answer: Thanks for the comments. In the revised manuscript, we have improved the method description by addressing comments raised by all reviewers. To keep the track of change, we keep the original latex style, which can be later transformed to be in MDPI template.
