**1. Introduction**

Developments in the invasive use of grid-flexibility options, such as demand-side management (DSM), require pliability in load prediction mechanisms to match temporal and spatial differences between energy demand and supply [1]. Recent advancements in DSMs, such as vehicle-to-grid (V2G) technologies and renewable energy policies, induce new perspectives for energy demand-supply imbalance management [2]. A critical factor in this is the reliable prediction mechanism of demand loads [3]. Thus, much attention has been given to demand load forecast mechanisms over the past decade [4]. A significant drawback of these prediction mechanisms is the lack of reliable high-sampled historical demand data, unscalable predictive models to match consumption patterns, and insufficient information on load prediction state uncertainties [5,6]. Energy consumption differs among load types, consumption behavior, and time of energy usage. For loads with routine energy consumption, such as office building loads, the energy sequence is stationary with minimal energy variations within the operation periods. However, for non-routine energy consumption or generation loads, such as hotels and renewable power, respectively, the energy sequence is randomized, with many variabilities in the energy profile patterns. Such a situation can pose a challenge to predict with generalized prediction models.

Demand load forecasting is an established yet still very active research area [7]. The literature on prediction models on energy consumption is enormous, with recent studies harnessing the power of machine learning (ML) to develop highly-generalized predictive models. Before this, classical predictive methods dwelled mainly on statistical analysis [8–10]. Consumption patterns were stable with fewer variations with these loads; hence, models such as support vector regression (SVR) and auto-regressive integrated moving average (ARIMA) were used for short-term prediction of loads [11–18]. The relevance of these methods is dependent on the extensive dataset with collinear measuring variables, which, in most cases, are difficult to come by. Therefore, classical predictive methods were, as a result, incapable of capturing random variations in the data patterns [19]. Efforts to improve the forecasting methods incorporating such diversity and unevenness prompted attempts to replace classical regression models with ML techniques [20]. Predominate among these techniques is the artificial neural network (ANN). The ANN is already known for its dominant utilization in energy forecasting methods [13,21,22]. However, because of its inherent complexity, it mostly leads to overfitting. So, an ensemble model that harnesses the merit of ANN potential, together with other ML algorithms, could be devised for high accuracy load prediction. Recently, Wang et al. proposed an ensemble forecasting method for the aggregated load with subprofile [23]. In this work, the load profile is clustered into subprofiles, and forecasting is conducted on each group profile. Apart from the fact that this algorithm is based on a fine-grained subprofile, which may not be readily accessible from every energy meter, cluster members with similar features but different load profiles are problematic to cluster. Hence, a centroid representation of a cluster may lead to higher variance in load prediction. In [24], Wang et al. proposed a combined probabilistic model for load forecast based on a constrained quantile regression averaging method. This method is based on an interval forecast instead of a point forecast. Apart from increasing computational time required for bootstrapping, much data is needed, and interval resolution may not be optimal for other data. Given this, this paper focuses on a predictive ensemble with limited available historical datasets to develop a scalable online predictive model for demand load forecasting. In this study, date meta-data parameters and weather condition variables serve as inputs to the proposed framework. As opposed to the models mentioned above, the proposed prediction model is not based on any specific data sampling interval or strict prediction interval. The model is adaptable to different distributed demand loads with a varying number of predictors at various sampling intervals. The contributions of this study are to:


The remainder of the paper is organized as follows. Section 2 discusses some challenges in demand load forecasting. Section 3 describes the probabilistic load forecasting model generation for stochastic demand load forecasting and error compensation methods. In Section 4, the results of the parametric models and the ensemble forecasting model on different case studies are presented.
