Load forecasting using a multivariate meta-learning system

Abstract

Although over a thousand scientific papers address the topic of load forecasting every year, only a few are dedicated to finding a general framework for load forecasting that improves the performance, without depending on the unique characteristics of a certain task such as geographical location. Meta-learning, a powerful approach for algorithm selection has so far been demonstrated only on univariate time-series forecasting. Multivariate timeseries forecasting is known to have better performance in load forecasting. In this paper we propose a meta-learning system for multivariate time-series forecasting as a general framework for load forecasting model selection. We show that a meta-learning system built on 65 load forecasting tasks returns lower forecasting error than 10 well-known forecasting algorithms on 4 load forecasting tasks for a recurrent real-life simulation. We introduce new metafeatures of fickleness, traversity, granularity and highest ACF. The meta-learning framework is parallelized, component-based and easily extendable.

Formation of prognostic software support for strategic decision-making in an organization

“…To date, there is a gap between the successful implementation of a particular forecasting method and the development of a complex prediction system [19] to satisfy the long-term needs for organizing prognostic support.…”

Cloud Based Metalearning System for Predictive Modeling of Biomedical Data

“…This methodology is used in fraud detection [36], time series forecasting [37], load forecasting [11], and others.…”

Assessment of an Adaptive Load Forecasting Methodology in a Smart Grid Demonstration Project

“…The difference could be due to the use of electrical heating equipment in Network 1 [30] and also because of the typical load profile in household customers in Spain, which corresponds to peak values at 19-22 h (evening) and at 11-14 h (morning). The load pattern of commercial customers is heavily influenced by opening and closing hours (9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21) and the use of cooling appliances during summer. …”

“…Hybrid load forecasting frameworks have been used lately, which combine both statistical approaches and artificial intelligence-based models [15]. In [16], a meta-learning system is proposed, which automatically selects the best load forecasting algorithm out of seven well-known options based on the similarity of the new samples with previously analyzed ones, considering not only univariate data, but also multivariate data. As an improvement to traditional forecasting algorithms, other novel hybridizations have been proposed lately, such as chaotic evolutionary algorithms hybridized with SVR [17] or least squares SVM with fuzzy time series and the global harmony search algorithm [18].…”

“…Although the design of a meta-learning forecasting algorithm such as is shown in [16] is out of the scope of the paper, some comparisons between adaptive and meta-learning techniques are summarized next:…”