The relatively unpredictable nature of cloud formation, disappearance and movement causes short-term variability of irradiance levels, resulting in variable solar PV power generation.
There are several ways to manage this variability and cloud forecasting using cloud prediction technology (CPT) is likely to be a part of the solution in both on- and off-grid applications.
On remote and isolated hybrid power stations that often lack the diversity of generation technologies, the rapid and often unexpected change in both generation and demand must be managed to ensure power fluctuations do not exceed the ramping capability of supporting technologies (e.g. diesel generators).
Operational constraints (e.g. curtailing PV) and/or additional technologies (e.g. battery) may be required to ensure the overall power system ramp rate remains within the limits needed to guarantee stability.
Several approaches to ramp rate control exist, and include:
Optimising the selection of available thermal generators based on CPT signals;
Pre-emptively curtailing solar PV generation based on CPT signals;
Using smoothing battery energy storage systems (BESS) to manage PV generation and demand fluctuations; and,
A hybrid approach that employs a mix of the above.
CPT systems have inherent forecasting uncertainty, and the control system designer must weigh the balance of both false positive predictions (i.e. predicted cloud cover events that do not come to pass) and false negative predictions (i.e. cloud cover events that the CPT failed to predict).
When applied to the specific application of ramp rate control, CPT false negatives may result in non-compliance with network technical requirements, while false positives merely result in temporary unnecessary curtailment of PV generation.
Therefore, a designer will tend to prefer a conservative integration of the CPT, resulting in a considerable number of false positives being admitted.
Whilst CPT is receiving increasing amounts of attention in the industry, both in the on- and off-grid space, the technology has not yet demonstrated sufficient technical reliability to deliver the economic benefits it promises, nor is it yet available as an off the shelf product; both necessary preconditions for widespread adoption.
To support the development of the technology, Ekistica recently evaluated the accuracy of forecasts from CPT systems deployed as trials (i.e. power station does not respond from CPT signals) in off-grid applications that utilise solar PV and diesel generation.
The results were presented by two scores:
Thereliability score measured successful prediction by the CPT system of cloud events that resulted in a significant drop in available PV generation, thereby allowing the risk of system outages to be avoided.
The efficiency score measured successful prediction by the CPT system of clear weather which would thereby provide opportunities for reduced diesel fuel consumption.
Since a lower reliability score would be associated with more system outages, while a lower efficiency score would merely result in reduced fuel savings, the reliability score is the primary determinant of the overall feasibility of CPT integration in off-grid applications.
However, a perfect reliability score could be achieved simply by forecasting a “cloud event” signal at all times. Therefore, the efficiency score is a necessary supporting point of evaluation.
Both the reliability score and the efficiency score are calculated in terms of significant drops and evaluation events. Roughly speaking, an evaluation event occurs at the very beginning of every period of cloudy weather that causes a significant drop in available PV generation.
Once the CPT system has correctly predicted this initial drop, it is presumed that the power station’s control system will remain in a conservative mode for an extended period, and the ongoing CPT forecast accuracy during this period need not be considered in the evaluation.
Some of the main definitions used to define significant drops and evaluation events include:
Drop threshold: The magnitude of a drop in available PV generation constituting a significant drop. E.g. this might be set to equal the available spinning reserve, as any drop smaller than this will be covered by the system, regardless of the CPT forecast.
Drop window duration: The maximum duration over which a drop in available PV generation can occur while still constituting a significant drop. E.g. the time required for a new diesel generator to become available from a cold-start.
Exclusion window duration: To be considered an evaluation event, a significant drop must not be preceded by any other significant drops within this time frame. E.g. this time might equal the diesel generator minimum run time. If the forecast prompts a change to the selection of generation units, then that change will remain in place for at least this long.
Minimum prediction lead time: The minimum lead time for prediction of an evaluation event to be considered successful. E.g. the time required for a new diesel generator to become available from a cold start. For a forecast to be useful, its predictions must be made with at least this much fore warning.95% confidence intervals for the reliability and efficiency score of forecasts from two CPT systems are shown inFigure 1.
The best performing CPT system (Vendor 2) correctly predicted every one of the 690 assessed evaluation events that caused a significant drop in PV generation, representing a level of reliability approaching the standard required for integration into off-grid power systems (i.e. minimal risk of power outage events).
The same forecast achieved an efficiency score of 30%, the lowest of all the forecasts from Vendor 2.
Theoretically, if this forecast were integrated into an off-grid power system using diesel generation, there would have been opportunity to save on diesel consumption 30% of the time during clear weather, providing relatively modest opportunities for fuel savings.
The other forecast (Vendor 1) was able to achieve higher efficiency scores, but only at the expense of failing some evaluation events, which would likely result in power outage events if used.
The apparent trade-off between reliability and efficiency scores reflects a more general compromise between true-negative and true-positive predictions.
The reliability/efficiency trade-off is particularly visible in Figure 1: As Vendor 2’s forecasts achieve lower efficiency scores they are progressively more sensitive, resulting in more of the evaluation events being successfully predicted, and a higher reliability score.
However, these incremental gains in reliability come at an increasingly high cost in the efficiency score.
Figure 1. Scatter plot showing both efficiency and reliability scores and their 95% confidence intervals, for two of the assessed CPT system forecasts Figure 1. Scatter plot showing both efficiency and reliability scores and their 95% confidence intervals, for two of the assessed CPT system forecasts
Developers seeking to incorporate CPT into ramp rate control systems in off-grid power systems should carefully weigh the risks and costs associated with false negative predictions.
While a hybrid approach to managing ramp rate can provide cost benefit solutions, the degree of this benefit depends heavily on the forecast reliability, and upon the frequency of false-negative predictions that can be admitted within the network rules.
The results from the assessment outlined above are promising and indicate that CPT forecasting accuracy might be approaching levels required for integration into off-grid power system applications. However, more trials and further analysis is required to confirm this and to quantify the overall economic benefit of integration.
These lessons may also be applicable to the on-grid space. On large interconnected grids, penalties may apply to generators for any power dispatched that is not in-line with forecasts.
Traditional generators self-forecast, whereas the responsibility for forecasting solar and wind generation falls to the Australian Energy Market Operator. A new evaluation framework could be developed to determine the capability of CPT systems to self-forecast for solar and wind generators and the impact this has on contribution factors to Frequency Control Ancillary Services.
Initial results from the Short-Term Forecasting Round initiative led by The Australian Renewable Energy Agency demonstrates that self-forecasting solar generation can be more accurate than forecasts provided by AEMO.
A consistent evaluation framework, like the one described above, may help in measuring this accuracy and determining the economic benefit for integrating CPT systems in on-grid applications.
