An Efficient Short Term Load Demand Forecasting Using a Novel Parallel CNN-BiLSTM Hybrid Neural Network for Bangladesh Perspective

Abstract

STLF (Short Term Load Forecasting) has traditionally become one of the most crucial, delicate, and precise demanding variables in energy systems. An efficient STLF enhances not only the financial feasibility of the system, but also its safety, consistency, and dependability in performance, allowing for the realization of a prospective Smart Electricity System. The state-of-the-art models exhibit significant nonlinearity in load information from available projections, as well as limited applicability in real-world circumstances. However, for real-world application, the energy forecasting area requires better resilience, improved prediction accuracy, and adaptability capacity. The study given in this paper supports the case for a hybrid strategy, in which the complimentary qualities of several cognitive methodologies are merged to provide a superior solution to the STLF problem. The deep learning models for STLF integrated with statistical techniques are presented in this paper for an accurate load forecasting. Temperature, humidity, and day type are all taken into account since they have a substantial effect on the overall performance of an appropriate STLF. The load demand data has been collected from the PGCB database, the weather data has been collected from the rp5 archive and the holidays are considered from the government calendar which excludes the data collection part of our research. The data refinement process has been done where many preprocessing techniques are applied on the raw data. With the proper data analysis and scaling the further process fed into the deep learning models where the training, validation and testing were done. In terms of computing complexity and prediction accuracy, the suggested model outperforms the prior hybrid models. The proposed technique of our methodology against existing power prediction information reveals that it performs better in terms of precision and accuracy. The evaluation has been done considering the Mean Absolute Percentage Error (MAPE) and R-squared score which outperforms the existing literature reviews. When compared to existing baseline models, the suggested technique had the lowest error rate on the Power Grid Company Bangladesh dataset.