Predictive Modelling of Load Dynamics in Smart Grids Using ECO-GRU Architecture and Exploratory Temporal Feature Engineering

Load Forecasting (LF) is essential in order to operate and plan the electricity system. Therefore, predicting future energy demands is crucial for controlling consumption by matching utility offers with consumer demand. To statement these problems, this paper proposed an ECO-Gated Recurrent Unit (ECO-GRU) to designed for time series load forecasting in smart grid. The raw data such as voltage, current, Photovoltaic (PV)/wind power, power etc. are as an input to pre-processing stage. Four techniques are utilized in the pre-processing stage time/lap alignment, re-sampling, handling missing values, and exploratory analysis. The time lap alignment and re-sampling utilized for standardizing timestamps and re-estimates the data. Also, handling missing values and Exploratory Data Analysis (EDA) utilized for finding missing values in the input data and analyse the presence of multiple hidden features. Feature engineering, a log feature and rolling statistics computes the statistical measures over rolling period of data for further analysis. A proposed classifier ECO-GRU utilized for reducing the training parameter and ensures the prediction accuracy. Implementing Python software, the proposed ECO-GRU accommodate the actual generation pattern better than existing methods and produce the Mean Absolute Error (MAE), and Root Mean Square Error (RMSE) , R2 value of 0.13, 1.8 and 99% respectively.