Electrical Load Short-term Forecast Using Feed-Forward Backpropagation Neural Network Methode

Penelitian ini, membahas perbandingan metode peramalan. Metode peramalan yang digunakan adalah Feed Forward Backpropagation Neural Network dan Generalized Regression Neural Network. Peramalan beban listrik jangka pendek mengacu pada perhitungan beban harian, temparatur udara harian. Peramalan ini akan meramalkan beban listrik pada tanggal 3 Mei 2007. Peramalan ini akan membandingkan metode Feed Forward Backpropagation Neural Network dengan berbagai fungsi pelatihannya dan metode Generalized Regression Neural Network, pada metode Feed Forward Backpropagation Neural Network hasil peramalan yang nilainya mendekati beban aktual adalah pada pukul 05.00 dengan hasil peramalan dari fungsi pelatihan traincgf sebesar 0,08% sedangkan pada metode Generalized Regression Neural Network pada pukul 17.00 dengan hasil peramalan sebesar 29,06% dan beban peramalan yang selisihnya paling besar dengan beban aktual pada metode Feed Forward Backpropagation Neural Network adalah pada pukul 21.00 dengan hasil dari peramalan fungsi pelatihan traincgp sebesar 55,14% sedangkan pada metode Generalized Regression Neural Network pada pukul 12.00 sebesar 4,34%. Peramalan terbaik terdapat pada metode Feed Forward Backpropagation Neural Network dengan fungsi pelatihan traincgf di mana nilai MAPE mendekati nilai beban real yaitu sebesar 2,52%.

This study discusses the comparison of forecasting methods. Forecasting methods used are Feed Forward Backpropagation Neural Network and Generalized Regression Neural Network. Short-term electricity load forecasting refers to the calculation of daily load, daily air temperature. This forecast will forecast the electricity load on May 3, 2007. This forecast will compare the Feed Forward Backpropagation Neural Network method with its various training functions and the Generalized Regression Neural Network method, the method of Feed Forward Backpropagation Neural Network forecasting results whose value is close to the actual load is at 05.00 with the traincgf training function of 75.6 MW or 0.08% while the Generalized Regression Neural Network method at 17.00 with forecasting results 29.06% and forecasting load which is the biggest difference with the actual load on the Feed Forward Backpropagation Neural Network method is at 21.00 with a traincgp training function of 55.14% while the Generalized Regression Neural method Network at 12.00 is 54.34%. The best forecast is found in the Feed Forward Backpropagation Neural Network method with the traincgf training function where the MAPE value approaches the real load value of 2.52%.