Interpretable Modeling for Short- and Medium-Term Electricity Demand Forecasting

Abstract

We consider the problem of short- and medium-term electricity demand forecasting by using past demand and daily weather forecast information. Conventionally, many researchers have directly applied regression analysis. However, interpreting the effect of weather on the demand is difficult with the existing methods. In this study, we build a statistical model that resolves this interpretation issue. A varying coefficient model with basis expansion is used to capture the nonlinear structure of the weather effect. This approach results in an interpretable model when the regression coefficients are nonnegative. To estimate the nonnegative regression coefficients, we employ nonnegative least squares. Three real data analyses show the practicality of our proposed statistical modeling. Two of them demonstrate good forecast accuracy and interpretability of our proposed method. In the third example, we investigate the effect of COVID-19 on electricity demand. The interpretation would help make strategies for energy-saving interventions and demand response.

Short-term power load forecasting based on the CEEMDAN-TCN-ESN model

“…The dropout layer is established to prevent overfitting, and the dropout rate is set to 0.2 (i.e., some neurons will be randomly deleted with a probability of 0.2). The expansion coefficient (dilations) is set to [ 1 , 2 , 4 , 8 ] (exponentially increasing by 2), and the number of filters is set to 64. Since temporal signal modeling cannot violate the temporal order, which will produce causal convolution, the padding parameter is set as ‘Causal’.…”

“…Dudek [ 1 ] simplified the univariate short-term electric load forecasting problem by filtering out trends and seasonal variations with longer than daily variation periods, using linear regression for local modeling in the domain of the current input data, and using the stepwise regression method and the Lasso Lars method to reduce the number of explanatory variables in the regression analysis. Hirose [ 2 ] used a variable coefficient model to capture the multivariate nonlinear structure in short-term load data while using the nonnegative least squares method to estimate nonnegative regression coefficients, achieving more accurate predictions. The research method proposed by Dudek and Kei provides a basis for the daily dispatch of power grids.…”

A Multivariate Time Series Analysis of Electrical Load Forecasting Based on a Hybrid Feature Selection Approach and Explainable Deep Learning

“…Basic structures, such as multi-layer perceptrons (MLPs), recurrent neural networks (RNNs), and convolutional neural networks (CNNs), along with their combinations, are contemporary deep learning architectures. For time series tasks, RNN offers several advantages, such as non-linear learning capabilities, effectiveness in obtaining information about the time series, and the ability to discern the impact of other parameters on the load data simultaneously [5]. The LSTM architecture was proposed to address the issue of vanishing gradient when processing long sequences in RNNs [6].…”

Design of a Framework for Demand Forecasting Using Time Series Decomposition-Based Approach

“…The higher the degree of interpretability, (that is, the degree to which humans can understand the basis of judgments made by machines), the easier it is for decision makers to understand the basis for specific predictions, and more proactive decision making becomes possible. Thus, for demand forecasting in a business environment, it is important to achieve highly interpretable forecasts in addition to accuracy (Wu, Wang, Tao & Zeng, 2023;Hirose, 2021). The decomposition process is fundamental to studying and exploring time series data (Kamath & Li, 2021).…”