A Comparative Analysis of Neural Network Based Short Term Load Forecast Models for Anomalous Days Load Prediction

Abstract

Load forecasting plays a very vital role for efficient and reliable operation of the power system. Often uncertainties significantly decrease the prediction accuracy of load forecasting which affect the operational cost dramatically. In this paper, comparison of Back Propagation (BP) and Levenberg Marquardt (LM) neural network (NN) forecast model for 24 hours ahead is presented. The impact of lagged load data, calendar events and weather variables on load demand are analyzed in order to select the best forecast model inputs. The mean absolute percentage errors (MAPE), Daily peak error and regression analysis of NN training are used to measure the NN performance. The Forecast results demonstrate that, LM based forecast model outperform than BP NN model for performance matrices. This model is used to predict the load of ISO-New England grid.

A Machine Learning Model Ensemble for Mixed Power Load Forecasting across Multiple Time Horizons

“…It has to be noted that p (t) values contain the current and past, average and difference measures of the AP values, p(t+s) is the output, i.e., the mixed power load s fifteen-minute intervals ahead, whereas w (t) components contain the respective weather-related inputs of cloud coverage, wind speed, humidity, and temperature, respectively. Current and past AP measures p (t−i) , i = 0, 95, 671 p (t−i) , i = 0, 4, 92, 668 p (t−i) , i = 0, 8, 88, 664 p (t−i) , i = 0, 12, 84, 660 p (t−i) , i = 0, 24, 72, 648 The choice of the particular set of input variables can be justified as follows: The fact that electric load time series presents a strong dependency on previous values [47] strengthens the selection of such input variables in the form of (a). Trying to capture the trend of electrical load, differences (b) between current and previous AP values are frequently employed [31].…”

“…Particular calendar indices are also used by [43,44], while temperature and humidity data by [45]. These variables are common in most models, as the positive impact of exogenous variables on load forecasting has been confirmed by a number of articles, e.g., [46][47][48][49]. In many cases, the Levenberg-Marquardt algorithm is selected for artificial neural network training showing a better performance over other algorithms [50,51].…”

A comparative analysis of artificial neural network architectures for building energy consumption forecasting

“…Hu et al 22 proposed an STLF model based on a generalized regression neural network (GRNN) and reported that the prediction accuracy of the model was higher than that of the backpropagation neural network (BPNN). Ertugrul 23 Since the ISO New England dataset used in previous studies [27][28][29] has extensive geographical coverage of the collected electric load, it showed uncomplicated electric energy consumption patterns. Therefore, ANN with one HL showed satisfactory prediction performance by training simple patterns adequately.…”

“…Persistence, support vector regression (SVR), single-HL feedforward NN, RVFL, EMD, and EMD-based SVR). Reddy 21 proposed a Bat algorithm-based backpropagation approach for forecasting short-term electric loads [27][28][29] has extensive geographical coverage of the collected electric load, it showed uncomplicated electric energy consumption patterns. Therefore, ANN with one HL showed satisfactory prediction performance by training simple patterns adequately.…”

“…16,17 To date, numerous artificial neural network (ANN)-based STLF models have been constructed to predict exact electric loads and have exhibited favorable performance. [18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35] To further improve their performance, it is critical to effectively tune their hyperparameters. All ANN models have various hyperparameters (i.e.…”