Abstract:
This study presents forecast of peak electrical load of the Electricity Generating Authority of Thailand (EGAT) and the electrical energy of the Metropolitan Electricity Authority (MEA). In addition, input data were clustered by K-means algorithms before training by Artificial Neural Networks (ANNS) and Adaptive Neuro- Fuzzy Inference System (ANFIS). In this study, the input data consists of historical peak load statistics of Thailand (simple moving average), historical electricity statistics of Bangkok (simple moving average), month codes, quarterly gross domestic product (QGDP) and cluster number. The results of training and testing show that ANN model with two hidden layers exhibits the most accurate performance for peak electrical load of the Electricity Generating Authority of Thailand (EGAT). Meanwhile, ANN model with four hidden layers provides the most accurate forecast on electrical energy of the Metropolitan Electricity Authority (MEA). Furthermore, this study found that input data which are clustered by K-means algorithms before training and testing have better accuracy than the non-clustered data. Moreover, in the prediction of the peak load of EGAT in 2015, 2016 and 2017, the ANFIS model shows more accurate forecasting results than ANN model with two hidden layers. Meanwhile, in the prediction of electrical energy of MEA in 2016, 2017 and 2018, the ANFIS model exhibits better performance than ANN model with four hidden layers. The above forecast models can be applied to improve electrical load management of both organizations to be more efficient, including, maintenance and procurement of electrical energy from private sector, forecasting monthly or quarterly profits, determine the area of solar power purchase and investment in new substations, etc.