Determining Optimal PV System Monitoring Granularity
Inside this Article
Stiff competition in PV development is driving the need to simultaneously reduce costs and increase performance of solar assets. EPC contractors and solar integrators must be able to design, procure and construct projects at lower total costs to earn business. Financers, seeking lower capital costs from investors, apply pressure to reduce production risk on their downstream EPC, integrator or operations partners. Once an afterthought, the selection of monitoring systems is getting more attention at earlier stages of project development because monitoring is an essential tool for maintaining, managing and optimizing long-term system performance and revenue.
Small improvements in performance can have an amplified effect on net profit, so visibility is crucial. While most PV sites are instrumented with a revenue-grade meter, this monitoring approach alone rarely provides sufficient performance data. With pressure to balance cost, performance and risk, can developers afford to install additional dc-side data acquisition systems to safeguard against unforeseen losses? Can they afford not to? Although dozens of variables and risk factors come into play, an elegantly simple model can help guide these decisions in an economic and rational way.
A Call to Action
Monitoring provides visibility, both locally and remotely, to energy production and its variations, and assists in qualifying the need for field repairs. At minimum, monitoring provides the hourly, daily, weekly, monthly and quarterly assurances of expected performance. When a deviation or underperformance occurs, monitoring provides the first critical step in a longer process of detection, analysis and action.
The best monitoring solutions in the market create a call to action. The sequence of detection-analysis-action involves many factors, such as measurement accuracy, appropriate analysis, effective reporting, weather, identification of a root cause or causes, prescribing repair tickets, procuring parts, and scheduling technicians and on-site procedures. More often than not, diagnosis and repair require a site visit by a service technician. This “truck roll” is costly. Owners and operators must carefully consider the appropriate level of information needed from each site to guide this decision. How important is it to avoid performance and revenue losses? What variability in performance is expected? What level of site instrumentation is ideal?
A Revenue-Grade Meter Is Not Enough
Commercial-grade monitoring solutions typically start with a site-level revenue-grade meter to measure the collective ac power generation at the site. These highly accurate meters can measure power output with ±0.2% accuracy. So what are the potential benefits of inverter-, zone-, string- and module-level monitoring? (See Figure 1.)
Think of the power flow from dc to inversion to ac as one connected path that is measured at multiple points. A typical monitoring setup has an ac revenue meter at ±0.2% accuracy and an irradiance meter at ±5% accuracy. Operators often compare readings from the two meters to judge system performance. However, the accuracy of such comparisons are at the mercy of the weakest link in the chain. Statistics on error propagation imply an error of greater than 5% when taking the difference of these two measurements to judge performance issues. You need better granularity to achieve higher accuracy.
We interviewed asset managers who stated they often do not react to site underperformance unless it is greater than 5%, or in some cases 10%, of the expected monthly or quarterly target. Why? The combination of irradiance variation, instrumentation noise, normal system variation and lack of granularity add up to inaction.
For example, Figures 2 and 3 illustrate data from a 350 kWac PV site in North America. The two plotted graphs include data from the same system on two consecutive days. Figure 2 shows these two plots set side by side. They look similar, and at a casual glance they probably would not cause concern. However, if overlaid as in Figure 3, these two seemingly normal production days reveal differences in production of approximately 5%.
In this example, a seemingly indiscernible difference represents 5% in production losses. For a 350 kWac system producing electricity at a value of $0.15/kWh over a month, this 5% loss is worth several hundred dollars, and over a year it is worth approximately $4,000. Would you accept such losses if you could easily prevent them at a fraction of the cost of the lost revenue?
A similar analysis includes irradiance measurements. Changes in irradiance are a dominant factor in minute-to-minute and day-to-day performance changes. Graphing irradiance and generation on the same plot is a common technique. It allows you to visually correlate changes in irradiance and power. If generation drops precipitously but follows a commensurate drop in irradiance, you can easily spot the cause. However, this approach does not detect soiling or partial shading on an array. A soiled or partially shaded array will also track with irradiance, but it will not produce as much energy as it could, as shown in Figure 4.
In systems that rely solely on a revenue meter for monitoring, there is no absolute standard with which to compare the revenue meter data. You do not know if you have a problem unless it is big, or enough time passes for trending analytics to spot problems, or if you have adequate site instrumentation—particularly on the dc side—to detect and identify the issues. Granular monitoring provides a reliable alternative to the uncertainty of normal variation and measurement accuracy.
What Is Granular Monitoring?
Most designers are familiar with PV instrumentation and sensors of various types: irradiance meters, revenue-grade meters, inverter electronics, current transducers (CTs), smart combiners and other sources of measurement. These components and others used in combination can create a granular monitoring system with short gaps in the measurement chain and small blind spots in power management. Granular monitoring allows measurement of small sections of a system, increasing the likelihood of detecting minor changes in performance (see Figure 5).
Inverter and ac subsystem issues are often cited in the industry as the leading sources of system failures and energy loss. In a paper published by the Electric Power Research Institute (EPRI) in 2010 (see Resources), these two subsystems combined accounted for as much as 60% of the failure events and 70% of the energy loss in PV systems. This makes sense, as these complex subsystems are on the ac output side of the system. Thus, any equipment failure shuts down energy production of the whole array, or potentially a significant part of it. For this reason alone, inverter monitoring is recommended as a minimum.
In addition, the increasing use of string inverters and microinverters for nonresidential applications brings with it the benefit of more granular array monitoring by simply collecting a data feed from these inverters. For example, if you deploy 25 20-kW 3-phase string inverters on a 500 kW array, you can monitor 25 zones if each inverter is connected to the monitoring gateway. If each inverter offers two monitored MPPT channels, you can measure a total of 50 10-kW zones. Collecting data from these inverters then becomes a very economical monitoring approach.
The growing use of microinverters and dc-dc optimizers provides system operators with module-level performance data. However, module-level monitoring as a standalone solution requires a significant change in the system’s inverter architecture. As a result, in this article we focus on dc-side granular measurements including inverter, zone (or subarray) and string monitoring, since these options are available for any type of PV site design.
Assuming inverter readings are available, the decision to use zone-, string- or even module-level monitoring has been a topic of great debate in the industry. Some integrators consider string monitoring a cheap insurance policy. Others cannot justify the practice because they do not see the payback. We conducted a customer survey that indicated that the use of string monitoring may be as low as 10% in North America. However, with an increasing emphasis on performance management coupled with lower monitoring costs, both zone and string monitoring make sense in more systems than might initially be apparent.
The Benefits of DC Granular Monitoring
Figure 6 summarizes a dozen potentially detectable issues on the dc side of a system, the corresponding benefits of granular monitoring and the generalized value of losses. These losses are common and well documented in the North American PV industry.
The power of granular monitoring lies in its ability to detect small changes in output that otherwise go undetected. Small losses in the 1%–15% range are often lost in the noise of measurement accuracy or the uncertainty of expected performance. For example, the irregularity in day 2 output shown in Figure 7 might easily be mistaken for variation in irradiance or normal variation in performance. Day 3 indicates a further drop in output, but that matches the drop in irradiance.
Hidden from view in Figure 7 is the loss of an entire string on day 2. In Figure 8 string monitoring detects the loss of string 3 on day 2, which continues through the next day. Simple string alarms would immediately identify this issue. Left alone, the system would continue to perform but at a lower output than it should. The power in granular monitoring lies in its ability to detect otherwise imperceptible losses.
Similarly, if you are looking at site-level monitoring information, you can only capture gradual reductions from soiling and module degradation over long periods. With granular monitoring of subsections of the dc array, you can detect gradual degradation on shorter time scales. Figure 9 illustrates a scenario with 0.1% degradation per day. At the site-aggregate level, this rate of change is imperceptible and could easily be attributed to normal variations in performance.
Viewed at the zone-monitoring level (35 kW groups versus 350 kW site-level monitoring), the degradation is 10 times more visible and easier to detect. (See Figure 10.)
Measurement accuracy and uncertainty in expected performance help explain why asset managers and site operators commonly do not react to losses of less than 5% or sometimes even less than 10%. With site-level instrumentation only (such as revenue-grade and irradiance meters), small losses are imperceptible and can too easily be associated with changes in irradiance, temperature or measurement noise. Normal losses go unnoticed until they stack up to big numbers. Granular monitoring can arrest these losses by measuring smaller blocks of power at the zone or string level.
The Analysis Challenge
Increasing the granularity of dc-side monitoring allows the site operator to detect many issues that would otherwise go unnoticed, but what is the appropriate and economically feasible level of dc-side monitoring? How granular should you go?
This is a difficult question to answer, but you can approach it in several ways. In a perfect world, you would have at your fingertips all the data required to develop Pareto distribution graphs and frequency-of-failure data on systems for all potential combinations of inverters, modules and other key system components. You would also know the types, amounts and frequency of soiling for a specific location and array configuration. You would know which installers and EPCs have the least problems. You would know the mean time to repair (MTTR) and average cost for each type of site failure. With this information, you would be able to forecast the expected failure types, rates and cost for each unique site over a period of 20-plus years. This bottom-up approach would be ideal, but this level of information to forecast emerging or unforeseen failures is clearly not available.
You can simplify this complex challenge by looking at the problem from a different point of view. By estimating the value of losses associated with uncertainty and considering the cost of monitoring and service trips, you can assess whether it makes sense to deploy a granular monitoring solution for a specific installation and determine the appropriate level of granularity.
Granular Monitoring Decision Guide Inputs
We propose a simplified decision guide model with six key inputs. Figure 11 illustrates these key inputs, which are readily available or can be estimated during project design. Envect and DECK Monitoring have developed an online calculator based on this decision guide model. You can also develop your own decision guide based on the following inputs and related information.
Value of electricity. This represents the annual average value of the site’s energy production ($/kWh). The number should combine energy payments and RECs, with adjustments if time-of-use (TOU) uplift, coincident peak, penalties or other elements apply.
For example, if prevailing rates are $0.15/kWh, and roughly 30% of yearly output will have TOU uplift to $0.25/kWh (due to PPA schedules or effective net-metering value), use a weighted average of $0.18 (70% x 0.15 + 30% x 0.25). If REC payments of $200/MWh also apply, then add $0.20 for a total of $0.38/kWh.
Insolation factor. This combines several elements (annual insolation, PV efficiency, loss factors, power inversion efficiency and others) into a single number. It estimates the annual ac energy produced by each kW of PV system equipment installed on the site, with units of kWh/kW. It sounds complex, but it is easy to determine—simply multiply site insolation by total system derate factor (usually 0.75 to 0.85).
To estimate this factor, look up the site location on a standard PV insolation map that shows kWh/kWp/yr values. For example, North Central California has a value of roughly 1,850. Applying a derate value of 0.8, the insolation factor is 1,480 kWh/kW, or approximately 1,500 kWh/kW.
You can also use PVWatts to determine the insolation factor estimate. For example, go online to nrel.gov/rredc/pvwatts, select Version 1, pick someplace like Sacramento on the map, enter a PV system size of 1 kW, adjust the derate factor and hit calculate. On the table under AC Energy (kWh), the year total at the bottom is the value you want. In this example, changing the derate factor to 0.8 results in the value of 1,455 kWh/yr, or approximately 1,500 kWh/kW for a 1 kW system installed at this location.
System size. This is the PV nameplate capacity in kWp (dc). If the system has a dc-ac ratio of well over 1.3, then consider using the total ac kWp rating of all inverters multiplied by 1.25.
Loss risk. This represents a qualitative assessment of shading, soiling, dc subsystem and module-level losses as shown in Figure 6. Do not consider large system failures such as ac subsystem or inverter-level failures, which site or inverter monitoring would detect. Pick one of two levels—typical or low. A typical or medium-risk site has some expected soiling, minor intermittent shading, susceptible quality modules or other risk factors. A site with low risk of losses has limited to no shading or soiling and is solidly constructed using top-tier modules.
Cost of service trip. This is the weighted typical cost of deployment to make nominal repairs on a site. If a truck roll is involved, this number might be several hundred dollars or more. If transportation involves an airplane, the cost is much higher. However, if the site has a local maintenance and repair team, the cost might be low. If you do not know this number, you should determine this value for a given site.
Cost of monitoring. This is the annual cost of monitoring hardware and software. Add anticipated service or maintenance costs for the monitoring system if you prefer. If a third-party company is providing the monitoring solution, obtain a quote for the site. If you use internal systems or custom SCADA, this input should be the true loaded cost with salaries and overhead, not just the incremental monitoring components.
Decision Guide Logic and Examples
Our simplified model starts with an estimate of the potential losses due to measurement accuracy and expected performance uncertainty, referred to as a blind spot. You can estimate this using the first four input variables, along with other site-specific design assumptions. You can assume inverter-level monitoring as the baseline, and then subtract the cost of monitoring and the cost of a service trip to determine the value that remains. By comparing the differences in this value among zone and string monitoring, you can quickly determine if one of these is a better option than inverter-level monitoring. This approach, while based on simplified assumptions, allows you to frame the decision in economic terms.
The following model logic is incorporated into an online calculator that is available by contacting Envect or DECK Monitoring. The three monitoring granularity scenarios presented here are based on the results generated by this calculator.
1. Calculate the annual value of the blind spot at different levels.
- Estimate the inverter-level annual blind spot by multiplying site size, insolation factor and electricity value together, and then multiply by 5%. If you have made universal monitoring system upgrades for higher-accuracy measurements, adjust the blind spot accordingly. The resulting value is typically in the thousands of dollars.
- Estimate risk as typical or low using the information presented in Figure 6 and personal experience. If you are unsure, select typical. If you have taken pre-cautions and are sure that the risk estimate is low, you can cut your estimate of exposure for the inverter-level blind spot in half.
2. Subtract monitoring and service trip costs.
- Reduce the value of the blind spots for zone and string monitoring by the annualized monitoring cost. If you would spend more on monitoring than the value of the blind spot, you need not proceed further.
- Reduce the value of the blind spot by at least one service trip. If you cannot afford to take action to fix the issue on-site, then you need not proceed further.
- Compare the cost-adjusted blind spots to determine which option is best. Figure 12 illustrates the concepts necessary to determine the blind spots and to compare the differences between choices. The dollar values of the blind spots are shown in blue. The monitoring and service trip costs, shown in red, offset these values.
- Do the potential savings offset the costs? If both zone and string monitoring return positive numbers, you have choices. Is Delta 1 greater than Delta 2 or vice versa?
4. Apply judgment.
- If one result clearly shines, your decision is easy. If the results offer similar benefits but the site has high visibility or the potential for higher risks, choose a more granular option such as string monitoring. If neither payoff looks compelling, then revert to inverter-level monitoring. Most important, if the estimates surprise you or differ from your previous experience, check your assumptions and consider why the difference might fit the specific circumstances.
As with any model or analysis, this decision guide is not absolute—it is only as good as the assumptions used. Despite its simplicity, it does offer an approach that can appropriately frame and make sense of the otherwise complex process of choosing the appropriate level of monitoring granularity.
Monitoring Granularity Scenarios
The following three examples utilize our decision guide model logic to illustrate the results for three scenarios.
SCENARIO 1: 500 KW DEPARTMENT STORE ARRAY
Suppose a California department store now under PPA for $0.20/kWh has experienced slight underperformance, and the integrator wants to know if more-granular monitoring might be worth the investment on its next site. A default configuration of inverter-level monitoring at 5% accuracy has a blind spot worth $7,500 per year.
Zone monitoring can cut the blind spots to $750, for a maximum potential benefit of $6,750. From that, subtract budgetary costs of $1,220 for the additional monitoring and one unplanned repair. You will see a plausible benefit of $5,530 per year. Zone monitoring looks promising under the defined rule set.
String monitoring could provide visibility down to the $50 level and offer an improvement of $7,450 over inverter-level monitoring. It also offers very fast detection of problems, better analytics and more guidance on repairs. After subtracting budgetary costs of $2,870, the potential net benefit of avoided loss equals $4,580. While also better than inverter-level estimates, this does not quite beat zone improvements, so the recommendation in this scenario is for zone monitoring.
SCENARIO 2: 250 KW GROCERY STORE ARRAY WITH LOW-VALUE ELECTRICITY
Now consider a grocery store down the road with a smaller 250 kW array. Due to different financial arrangements, the value of the electricity ($0.10/kWh) is less than that in Scenario 1. The array is well designed and has low loss risk.
In this case, the decision guide process tends to rule out options. The potential blind spot is approximately 25% of the figure in Scenario 1, or roughly $1,880. However, the site’s low risk cuts the inverter-level exposure value by 50% for a value of $940. Zone monitoring would reduce the grocery store’s blind spot to $190, an improvement of $750. But the investment in zone monitoring, plus a budget for repair of $1,220, would consume all that benefit and more. Likewise, the numbers for string-level monitoring do not hold up. In this case, the model recommends inverter-level monitoring.
SCENARIO 3: 250 KW GROCERY STORE ARRAY WITH HIGH-VALUE ELECTRICITY
Now suppose that a similar grocery store has a much higher electricity value of $0.40/kWh due to legacy performance-based incentives. A qualified electrician who can perform any necessary repairs at a low cost is just a few doors away. Unfortunately, the site has issues with birds, intermittent shading, plant debris and other risk factors. The blind spot of inverter-level monitoring is $7,500 per year. Walking through the numbers, you can see that zone monitoring reduces the blind spot to $750 and offers $5,680 of potential net benefit.
Now consider string-level monitoring, which allows you to cut blind spots into small $100 pieces. After the $1,440 allowance for monitoring and incremental repair, the net potential benefit would be $5,960 per year. In this case, zone- and string-level monitoring granularity are both more favorable economically than inverter-level monitoring, with string monitoring at a slight advantage. An additional benefit of string-level monitoring is that it reduces repair time because technicians know which section of the array needs servicing before they arrive on-site.
These three scenarios reveal that changes in just one or two inputs can swing the recommendation from one end of the granularity spectrum to the other. We have modeled hundreds of combinations of inputs and site factors, and have drawn the following conclusions:
- Inverter monitoring is very affordable and should be deployed on every site.
- Zone and string monitoring can greatly reduce the blind spot but come at a price, including the cost of monitoring and the service trips to fix the problem and reduce the revenue losses.
- Zone monitoring is often the preferred choice to cost-effectively reduce the blind spot in nominal situations.
- String monitoring is recommended where the value of electricity is high, risks are typical or higher, and the site is serviceable at a low cost.
Granular monitoring is an effective tool for detecting energy and revenue losses that might otherwise go unnoticed due to measurement error or uncertainty concerning expected performance. These small losses can quickly add up, significantly affecting system energy production and revenue. The simple decision guide model that we have outlined allows the system designer to apply basic principles of estimated loss and costs to determine if increasing levels of granular monitoring are cost effective. This approach is also well suited for ruling out choices, such as when the cost of monitoring and a service trip exceeds the value of energy losses. Our model, when combined with good judgment, should help the designer understand the economics at play and simplify an otherwise complex decision.
Special thanks to DECK Monitoring for providing the raw site data used to develop the graphs presented in this article. In cooperation with DECK Monitoring, we conducted more than 20 interviews of asset managers, site owners, installers and operators during the article’s development. This article also includes information presented at the PV Power Plants 2012 and Solar Power International 2012 conferences.
Casey Miller / Envect / Bend, OR / envect.com
Todd Miklos / Envect / Fort Collins, CO / envect.com
Addressing Solar Photovoltaic Operations and Maintenance Challenges, EPRI, July 2010