Tailored Algorithms for Anomaly Detection in Photovoltaic Systems

Round 1

Reviewer 1 Report

The authors present a set of algorithms to detect what they call anomalies in PV system operation. The approach of the algorithms, which are based on the application/analysis of just PV output power timeseries is interesting, as no meteorological or other data is required, meaning that these algorithms could be broadly applied. Some of the algorithms, and assumptions that go with them could be documented better. Furthermore, the introduction should do a better job reviewing and discussing methods for anomaly detection in more detail. Please find my comments below.

L32: 'and thereby contributing': should be either "and thereby is contributing" or "and thereby contributes" 

L36: "greener" is a vague/subjective term, please use more accurate/appropriate wording

L37, "yielding": the goal of the target is to limit greenhouse gas emissions to such a level that the temperature increase does not exceed 1.5 degrees. I suggest rewriting to more accurately reflect this. 

L54-57: I understand that reference [20] might not be able to identify the cause of the anomalous output power, but the sentence "and thus unsuitable to identify" does not necessarily follow from the preceding text. Maybe something like "and is not designed to identify and distinguish" would be better

L84, Table 1: I think this is hardly a complete overview of PV system faults. For instance, component failure could also lead to reduced non-zero production (e.g. failure of single PV modules). MPPT problems could also be due to module mismatch. 

L95, Table 2: This is also not complete (e.g. vandalism/module theft, snow cover). Furthermore, it is not clear how high wind speed could directly lead to increase/decrease in power production. Also, it is not clear to what kind of case the shown percentages refer to, e.g. are these typical loss values, maximum loss values, average loss values in a large review? Also, what is the reference to which these losses refer? STC operation, nominal operation? For instance the 15% decrease for temperature could be higher for c-Si operating at 80 degC compared to STC (> 20% loss). 

L107-108, eqn (1): equations are not aligned

L112, offset: why did you define this offset at 2.5 hours? It seems quite a long period. Did you investigate the sensitivity of your algorithms to changes in this parameter? 

L120: Power has unit (k)W, not kWh (which is energy)

L131: How did you define this value of 0.85? Is the algorithm sensitive to changes of this parameter?

L139-152, Daytime shading algorithm: From this definition of the algorithm it becomes clear that during overcast situation with, intermittent solar irradiation, numerous local minima will exist, and thus many false positives will be generated. This is also shown in the results. This brings me also to the distinction you make between adverse and favourable weather: would it not be best if the algorithms first perform a check whether the weather is "favourable" (rather clear-sky) or not, before checking for e.g. daytime shading? 

L170-174, eqns (4a)-(5b), the formatting of this equations is not very clear, e.g. the parameters with +/- the offset are not defined as such

L175: why/how is the value defined at 40% of the slope? Is the algorithm sensitive to changes in this parameter?

L177: Why is suboptimal orientation considered an anomaly? Isn't it rather a given for e.g. the roof where the system is installed? And also, why then is an analysis of suboptimal inclination not included?

L214, eqn (8), why is the magnitude defined as the maximum shading loss, and not e.g. by calculating/estimating the loss in terms of energy generated? There could be very different losses in terms of energy for two systems with the same defined shading magnitude, even if shading length is the same...

L218, eqn (9): why is the shading loss time defined as such, and not as e.g. tmax2 - tmax1?

L267, Figure 4 caption: Typo in "(A) Somplete", should be "(A) Complete"?

L270: why is this exceedingly long?

L279, "spurious": wouldn't "false" or "erroneous" be better?

L442-443: why is the lag value different (5 hours) to the lead up value (2 hours)? Isn't a lag of 5 hours much too long? 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Tailored algorithms for the detection of several PV system anomalies (including brief and sustained daytime zero-production, daytime and sunrise/sunset shading, low maximum production and suboptimal orientation) are proposed in the paper. The work is well structured and the methodologies are discussed, as well as resuts are included
properly. However, some aspects should be revised/clarified:

• 
  A point highlighted by the auhtors is that '...that solely require the PV system production as input data...'(Section 1). In my opinion, current PV installations usually offer additional data mainly related to environmental parameters. This fact is an advantage of current solutions and they should be considered for estimations.

• - Different algorithms are proposed, but they are not compared to other contributions. Please, justifiy this aspect.

• - '...We distinguished two major types of PV system anomalies...' (Section 2) Please, justify this categorization.

• - Table 1. Please, update this table accordingly. - Table 2. This Table is elaborated by the authors? Please, include reference if required.

• - Section 3. '...We developed five algorithms for anomaly detection...' Please, justify this sentence.

• - Section 3.1 '...assume that time-series with daytime zero-production have at least one daytime observation where power generation, P day, i, is sufficiently close to zero...' This aspect depends on the inverter performance, and it can be varied depending on the commercial solution. Please, justify this assumption.

• - Section 3.4 '..., such that weather conditions (e.g., daytime period, air temperature) are regionally similar...' This assumption is very relevant and it should be clarified at the beginning of the paper.

• - Section 4.3 Discussion. In my opnion it is not clear the results and the false positives detected by the methodology (Table 5), which could be probably decreased by including environmental variables. Please, justify this point.

• - Please, compare the results to other contributions.

• - Finally, English grammar also needs checking in detail. Please, revise the paper. In my opinion, the contributions of the paper are poor for publication in Energies and the paper should be revised in detail before submission.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I'd like to thank the authors for their corrections. I now suggest to accept the paper