Forecasting Solar Power Generation: Revise the Forecast when it requires
Forecasting Solar Power Generation: Revise the Forecast when it requires
– Abhik Kumar Das
Del2infintity Energy Consulting Pvt. Ltd.
Forecasting and Scheduling (F&S) of solar power generation is an essential requirement of sustainable energy mix for the stable grid system since the variability and unpredictability inherent to solar create a threat to grid reliability due to balancing challenge in load and generation [1]. The integration of significant solar energy into the existing supply system is a challenge for large scale renewable energy penetration; hence the day-ahead and short-term renewable energy forecasting is needed to effectively integrate renewable power to the grid.
The most usable and conventional strategies in F&S of solar power generation are predicting the weather parameters using of NWP (Numerical Weather Prediction) models and converting the values of weather parameters into power generation using thesolar plant characteristics.Since the ensemble of daily solar power generation shows a partial periodic nature, the use of ANN (Artificial Neural Network) based forecasting strategies using NWP models is a widely accepted methodology in the research community. The recent advancement of different architectures of DNN (Deep Neural Network) has created a huge scope in forecasting solar power generation with high accuracy.
Considering the uncertainty of different variables, a good forecast does not produce a deterministic solution, but produces different scenarios having different probability values and the scenario with optimum probability can be considered minimizing the penalty due to the deviation. Hence, fromtheoretical considerations, F&S of solar power generation is best considered as the study of the temporal evolution of probability distributions associated with variables in the power generation.
Considering the regulations by CERC, FOR and other proposed regulations of different states of India, it is mandatory requirements of power generators to submit the day-ahead forecast of the power generation and if the error is more than a specified limit, there exists a penalty due to the deviation. For solar power forecasting 8 Intradayrevisions (one revision in every 1 hour 30 minutes) are allowed and this revision is effective from 4th time-block (1 time block = 15 minutes) onwards. It is interesting to see that the revision is not mandatory, but revision is permissible to tune the forecast accuracy.
Since solar power is variable in nature, the intraday revision in solar power is useful when there is an expectation of unscheduled fluctuations due to which power generation shows ramping behavior. The use of proper AI (artificial intelligence) techniques using DNN and proper plant specific localized solutions of NWP modelsmust be capable of forecasting not only solar power generation, but to decide when the intraday revision in forecasting is required by predicting and analyzing thesolar power ramping events.
A good forecasting methodology is not only a solution having the high forecast accuracy and low penalty due to the deviation, but a solution to maintain the similar or better accuracy in the minimum number of intra-day revisions. Multiple revisions are useful when those are necessary, but unnecessary submission of an intraday revision in every one or two hours proves the incapability of having proper pattern recognition techniques in forecast models.
When same or better forecasting solution is possible in lesser number of revisions, providing intraday revision in every one or two hours can increase the indirect cost of power generators; and if this revision requires real time data availability, multiple revisionsare not an economic viable option for small capacity solar plants.
Though there is an argument that intraday revision increases the forecasting accuracy, but the proper generation of patterns using DNN optimizes the number of revisions if the expectation of unscheduled fluctuations is low. Intraday revisions are useful only when there is high possibility of ramping events or the forecasting is using some simple nonlinear models.Hence power generators must utilize the ‘revision strategy’ intelligently using better forecast models which give similar or better accuracy in the minimum number of intraday revisions.
Reference:
[1] Das Abhik Kumar, “Quantifying photovoltaic power variability using Lorenz curve”, Journal of Renewable and Sustainable Energy, AIP, Vol.6 (3), June 2014
[2] Das Abhik Kumar, ‘Forecasting and Scheduling of Wind and Solar Power generation in India’, NTPC’s Third International Technology Summit ‘Global Energy Technology Summit’ 2016
[3] Das Abhik Kumar,‘Applicability of Error Limit in Forecasting & Scheduling of Wind and Solar Power in India’, Technical Paper ISGW 2017, New Delhi (accepted)
[4] Das Abhik Kumar et a.l “An Empirical Model for Ramp Analysis of Utility-Scale Solar PV Power”, , Solar Energy, Elsevier, vol. 107, pp. 44-49, September 2014.
