Abstract

For the past two decades, much of the research has been performed on Electrical Load Forecasting methodologies (ELF) over the world. The ELF is an important topic for energy management because of its significance in several federal policy related sectors, such as generation planning, appropriate expansion of electricity transmission and distribution networks in respective regions within time-frame, and suitable allocation of energy resources within masses [1]. The planning institutions of power utilities in developing and under-developing countries are making use of traditional statistical methodologies for electrical load forecasting purpose and are solely relying on these outdated techniques. Such obsolete methodologies are not capable of incorporating system non-linearities effectively, which are induced by the non-linear seasonal data. The modern-day machine-learning and deep-learning based non-linear time-series parametric modeling techniques are more suitable to handle the system dynamics and non-linearities effectively, rather than traditionally employed statistical methodologies [2]. However, the electrical load forecasting mechanism requires adequate modeling to handle such diversified systematic dynamic effectively, which can be accomplished using Exploratory Data Analysis (EDA). The aforementioned analysis highlights climatic and temporal factors as the potential input features for modeling electrical load forecasting methodologies [3]. As far as the energy demand is concerned, the annual report of International Energy Agency (IEA) reveals the fastest growth of annual electricity demand in developing countries [4]. The main contributing factors in this exponential growth are rapid urbanization and rising industrial revolution [5], [6]. Consequently, such diversified electricity demand should be forecasted using high-performance and robust forecasting methods for the appropriate electrical power generation, transmission and utilization planning. Therefore, the motivation behind this study is to encourage the planning authorities of electric utilities of developing countries with state-of-the-art linear statistical based parametric and non-linear machine learning and deep-learning based parametric methodologies. In this research, real-time meteorological and electrical load data of the city of Lahore in Pakistan in used to analyze the proficiency and reliability of several state-of-the-art linear statistical based parametric models and non-linear machine-learning and deep-learning based parametric methodologies. Based on performance indices, such as Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), R-Square value, and standard deviation, a comprehensive quantitative and qualitative analysis comparison has performed among these scientific rationales.