Machine Learning-Based Hybrid Algorithms to Forecast Power Generated by Solar Projects

The breakthrough allows day-ahead forecast of power generated by solar PV projects

July 31, 2021

thumbnail

Researchers led by Dr. Kalop of Urban Environmental Engineering and Professor Heo Jong-wan from Incheon National University, South Korea, have developed two new machine learning-based models that forecast the power generated by solar projects. Called the adaptive neuro-fuzzy inference system (ANFIS), the advanced models incorporating artificial intelligence efficiently forecast the power generated by photovoltaic systems ahead of time by up to a full day.

The study was published in Renewable and Sustainable Energy Reviews.

Integrating solar photovoltaic (PV) power into existing power grids is a challenging process as the power output of PV systems depends heavily on environmental factors. An accurate forecast of solar PV power generation is necessary for efficient integration of power into the existing power grids, noted the researchers.

The researchers have integrated two models – ANFIS and particle swarm optimization (PSO) with adaptive and time-varying acceleration coefficients.

The researchers explained that the two models are ‘hybrid algorithms’ because they combine a novel hybrid approach of adaptive swarm intelligence techniques and ANFIS in forecasting the power generation of solar PV projects at different time horizons, from 0 to 24 hours.

The two models – ANFIS-APSO (ANFIS-PSO with adaptive acceleration coefficients) and ANFIS-IPSO (ANFIS-PSO with time-varying acceleration coefficients) were developed to assist the program.

The performance of the proposed models was compared with other hybrid ANFIS models – ANFIS-PSO (ANFIS coupled with PSO), ANFIS-BBO (ANFIS coupled with biogeography-based optimization), ANFIS-GA (ANFIS coupled with a genetic algorithm), and ANFIS-GWO (ANFIS coupled with grey wolf optimization).

For this purpose, the climatic variables and historical PV power data of a 960 kW grid-connected PV system in south Italy were used to design and evaluate the models. Several statistical analyses were implemented to evaluate the accuracy of the proposed models and assess the impact of variables that affects the PV power values. The experimental results show that the proposed ANFIS-APSO attained the most accurate forecast of the PV power with R2 = 0.835 and 0.657, RMSE = 0.088 kW and 0.081 kW, and MAE = 0.077 kW and 0.079 kW in the testing phase at time horizons 12 hours and 24 hours, respectively.

Based on the results, it was observed that the newly constructed ANFIS-APSO outperformed the standard ANFIS-PSO model, including other hybrid models. The findings revealed that the model could be a potential new alternative to assist engineers in forecasting the power generation of solar projects at short and long-time horizons.

The researchers explained that the two models are ‘hybrid algorithms’ because they combine a novel hybrid approach of adaptive swarm intelligence techniques and ANFIS in estimating the power of PV systems at different time horizons, from 0 to 24 hours.

Recently a team from Monash University’s Grid Innovation Hub, Worley, and Palisade Energy Ltd also conducted a joint study to precisely predict wind and solar power generation using machine learning technology. The five-minute ahead forecast mechanism is expected to reduce the frequency of poor dispatch and support a higher share of renewables in the market without compromising on the overall stability of the grid.

Accurate forecasting of power generation is a tough task for renewable energy project developers. But the utilities insist on accurate forecasting and scheduling of renewable generation to efficiently integrate them into the grid. Technologies that enable efficient forecasting of renewables are in great demand both by the utilities and renewable energy generators.

RELATED POSTS