Hybrid Irradiation Forecasting Method Using Neural Network To Reduce Exponential Smoothing Error

Abstract

Now a days talking about renewable energy is getting more common as long as researches are trying to come up with new ideas and updating previous approaches. One of the most common way to study and planning for increasing efficiency is forecasting demand and available resources. For forecasting different variables, different methods experimented. However, none of the approaches could minimize the forecasting error to the suitable point in renewable energy field such as, irradiation and wind prediction. In this paper one of the most frequent method for forecasting solar radiation were studied and examined which is exponential smoothing and the result of calculation were trained by neural network (feedforward). Neural network always used as an approach for forecasting separately but in this paper it is going to be used as part of the main calculation. Meaning to say that, neural network can be combined any method which in this case is exponential smoothing. The results was noticeable improved. Normally the average exponential smoothing errors would be around ten percent based on calculated results. But after applying neural network for completing the process, average error was declined to six percent.

Smoothing in Neural Network for Univariat Time Series Data Forecasting

“…This research will combine two methods, namely ES and NN. Several other studies that are also of the same type include forecasting electrical loads where the combination of the two can produce better accuracy (Sulandari, Subanar, Suhartono, & Utami, 2016), (Mohammed, Bahadoorsingh, Ramsamooj, & Sharma, 2017), the application of NN to ES can reduce the error forecasting results by up to 6 percent (Parsi, 2016), TES-F provides better accuracy than forecasting with TES alone (Fajriyah et al, 2019), The combination of ES NN in predicting the number of broadband users in Indonesia can significantly improve accuracy than using only one method (Gunaryati et al, 2019), forecasting model produced by combining the ES-NN provides high accuracy (Smyl, 2020), and it turns out that the smoothing that has been done on the data can improve the performance of NN training (Muhamad & Din, 2016).…”