Daily prediction of solar power generation based on weather forecast information in Korea

 

Abstract:

 

Solar panel photovoltaic (PV) systems are widely used in Korea to generate solar energy, which is one of the most promising renewable energy sources. With regard to solar electricity providers and a grid operator, it is critical to accurately predict solar power generation for supply–demand planning in an electrical grid, which directly affects their profit. This prediction is, however, a challenging task because solar power generation is weather dependent and uncontrollable. In this study, a daily prediction model based on the weather forecast information for solar power generation is proposed. In the case of the proposed model, the cloud and temperature data available from the weather forecast information is used to predict the amount of solar radiation as well as a loss adjustment factor to reflect the possible loss of power generation due to the degradation or failure of the PV module. Using the proposed model, solar power generation for the following day can be predicted. The proposed model is embedded into a solar PV monitoring system that is commercially used in Korea, and it is shown to perform better than the existing prediction models.

 

Existing System:

 

In the literature, several recent studies on the short-term prediction of solar power generation have been reported. An adaptive linear time-series model was suggested to predict the hourly values of solar power for horizons of up to 36 h using 15- min observations of solar power from photovoltaic (PV) systems and the global irradiance forecast information provided every 3 h by the Danish Meteorological Institute as input data [1]. Four artificial neural networks (ANNs) were developed for 24-h-ahead forecasting of solar power generation of a 20-kW PV system using vapor pressure, humidity, cloud coverage, sunshine duration, temperature, and radiation as input parameters [2]. ANNs were also used to predict short- and mid-term solar power generation and to determine the time horizon with the highest representative for the prediction of the electricity generated by small-scale (750-W) solar power systems [3]. A 3-h-ahead prediction model was developed using a weather forecast metric of the sky condition [4]. As a part of ongoing work of [4], machine learning based on the support vector machine (SVM) was applied to the 3-h-ahead prediction using the weather forecast of six weather metrics provided every hour by the National Weather Service in the USA as input data [5]. A least-square SVM-based model was proposed for the 1-h-ahead prediction for the atmospheric transmissivity, which is converted to solar power according to the latitude of the site and the time of the day [6]. Fuzzy decision trees were developed to predict the present solar power generation using the present weather forecast information, past weather forecast data, and observed weather records [7]. The applicability of heterogeneous regression algorithms was investigated for 6-h-ahead solar power availability forecasting using the historical weather data including temperature, wind speed and direction, and humidity obtained from Rockhampton, Australia [8]. Recently, a methodology based on the publicly available numerical weather prediction models was presented to generate day-ahead power output forecasts for two PV plants in the American Southwest [9].

 

Proposed System:

 

In this study, we develop a daily prediction model for solar power generation. The prediction model is used for implementing a daily prediction software module that is embedded into the PV monitoring system. Fig. 1 depicts a typical network and hardware configuration of the PV monitoring system used in the Korea solar PV industry. As shown in this figure, the monitoring system communicates with the inverter and the sensor using the RS-232C cable, and it is connected to the server of the Korea Meteorological Administration (KMA) via the Internet. The KMA offers a weather forecast web service in Korea. In particular, it forecasts weather for a small area (such as a village or a town) in 3-h units over the following 48 h; such a type of forecast is called neighbourhood weather forecast. The weather forecast information includes 12 types of weather elements, a few of which include the temperature, humidity, precipitation probability, wind speed and direction, and sky condition. The really simple syndication or rich site summary (RSS) service is open to the public to forecast weather at 3-h intervals over the next 2 days, as well as to facilitate weekly weather forecast at a 1-day interval in an XML data format. The advantage of the RSS is that people can easily obtain an updated weather forecast without visiting the national weather service website as well as customise the obtained information according to their use.

 

Conclusion:

 

Most of the previous studies have focused on developing the regression model and the ANN model for predicting short-term solar power generation. These models have strong theoretical foundations but in practice they leave a lot to be desired. The regression model has to be regenerated periodically to fit the reality, and the ANN model is not easy to use without theoretical knowledge. Furthermore, both of them require a large amount of historical operational data to build prediction models. To overcome these drawbacks of the existing models, we developed a new prediction model for predicting a daily solar power generation in this study. The proposed model is based on the solar PV output formula with a correction factor and uses weather forecast information to estimate insolation and the correction factor. We performed the empirical experiment using the 27.34-kW solar PV system installed in GERI, Korea, to confirm that the proposed model works better than the existing models. The major advantages of the proposed model are five folds. First, it is easy to use because parameter tuning is done automatically without human intervention. Second, it does not require a large amount of historical operational data to build a model. Third, its computational logic is very simple, so it can be easily implemented in practice. Forth, its prediction accuracy is superior to competing models. Finally, and most importantly, it can reflect the performance degradation of the system due to aging or external environment properly by adjusting the loss correction factor accordingly.

 

References:

 

[1] Bacher, P., Madsen, H., Nielsen, H.A.: ‘Online short-term solar power forecasting’, Sol. Energy, 2009, 83, (10), pp. 1772–1783

 

[2] Chaouachi, A., Kamel, R.M., Ichikawa, R., et al.: ‘Neural network ensemblebased solar power generation short-term forecasting’, Int. J. Inf. Math. Sci., 2009, 5, (4), pp. 332–337

 

[3] İzgi, E., Öztopal, A., Yerli, B., et al.: ‘Short-mid-term solar power prediction by using artificial neural networks’, Sol. Energy, 2012, 86, (2), pp. 725–733

 

[4] Sharma, N., Gummeson, J., Irwin, D., et al.: ‘Cloudy computing: leveraging weather forecasts in energy harvesting sensor systems’. Proc. of the 2010 7th Annual IEEE Communications Society Conf. on Sensor Mesh and Ad Hoc Communications and Networks, Boston, Massachusetts, USA, June 2010, pp. 1–9

 

[5] Sharma, N., Sharma, P., Irwin, D., et al.: ‘Predicting solar generation from weather forecasts using machine learning’. Proc. of the 2011 IEEE Int. Conf. on Smart Grid Communications, Brussels, Belgium, October 2011, pp. 528– 533

 

[6] Zeng, J., Qiao, W.: ‘Short-term solar power prediction using a support vector machine’, Renew. Energy, 2013, 52, pp. 118–127

 

[7] Detyniecki, M., Marsala, C., Krishnan, A., et al.: ‘Weather-based solar energy prediction’. Proc. of the 2012 IEEE Congress on Computational Intelligence, Brisbane, Australia, June 2012, pp. 587–593

 

[8] Hossain, M.R., Oo, A., Ali, A.: ‘Hybrid prediction method for solar power using different computational intelligence algorithms’, Smart Grid Renew. Energy, 2013, 4, (1), pp. 76–87

 

[9] Larson, D.P., Nonnenmacher, L., Coimbra, C.F.M.: ‘Day-ahead forecasting of solar power output from photovoltaic plants in the American Southwest’, Renew. Energy, 2016, 91, pp. 11–20

 

[10] Forero, N., Hernandez, J., Gordillo, G.: ‘Development of a monitoring system for a PV solar plant’, Energy Convers. Manag., 2006, 47, (15-16), pp. 2329– 2336