A preview of the PDF is not available
Electrical Energy Consumption Forecasting for Efficient Energy Management in Smart Grid
Accurate, fast, and stable electrical energy consumption forecasting plays a vital role in decision making, energy management, effective planning, reliable and secure power system operation. Inaccurate forecasting can lead to the electricity shortage, wastage of energy resources, power outage, and in the worst case, power grid collapse. Contrarily, accurate forecasting enables policymakers and public agencies to make real-time decisions imperative for the energy management and power system’s secure and reliable operation. However, accurate, fast, and stable forecasting is challenging due to consumers’ uncertain and intermittent electrical energy consumption behavior. In this context, a rigid forecasting model with assertive stochastic and non-linear behavior capturing abilities is needed. Thus, several forecasting strategies have been emerged in state-of-the-art work, starting from conventional time series to modern data analytic methods to solve the non-linear electrical consumption prediction problems. The individual techniques partially resolved the forecasting problem by improving forecast accuracy. However, the improvement in accuracy is not up to the mark. Besides, individual techniques (conventional or modern) suffer from their inherent limitations. Due to inherent limitations, forecasting results of individual methods (conventional or modern) are no longer as accurate as required. To solve such problems, hybrid models developed, which fully utilize individual methods’ advantages and have comparatively improved performance. Only some models are commendable that improve accuracy, while others perform better in convergence rate. However, considering only one aspect (accuracy or convergence rate) is insufficient. Thus, accuracy and convergence rate both are of prime importance and can be improved simultaneously. Therefore, scholars and industries have the primary goal of developing a forecasting model, which provides robust, stable, and accurate electrical energy consumption forecasting for efficient energy management. With this motivation, a novel two-stage hybrid model is developed by integrating the electrical energy consumption forecasting stage with the energy management stage. The first stage is for electrical energy consumption forecasting, and the second stage is for efficient energy management. The first stage composed of four modules: (i) a novel cascaded framework based on factored conditional deep belief network (FCDBN), (ii) deep learning based forecaster cascaded with a heuristic algorithm based optimizer framework, (iii) support vector machine (SVM) based forecaster integrated with modified enhanced differential evolution (mEDE) algorithm framework, and (iv) accurate and fast converging deep learning based forecaster framework. The first module framework consists of data preprocessing phase, FCDBN training phase, and FCDBN based forecasting phase. The first module framework is developed in a cascaded manner, where each former phase’s output is fed into the later phase as input, namely cascaded framework. The second module is a hybrid framework composed of data preprocessing and feature selection, training and forecasting, and optimization phases. The third module is an integrated framework of data preprocessing and features engineering, SVM based forecaster, and mEDE based optimizer, namely FA-HELF. The fourth module is a cascaded framework of feature selector, FCRBM based forecaster, and GWDO based optimizer, namely FS-FCRBM-GWDO. The purpose of the modules in the first stage is to provide fast, accurate, and stable electrical energy consumption forecasting. To accomplish this goal first, the data is preprocessed to convert it into a usable format. Secondly, clean and prepared data is passed through feature selection and extraction phases to select the most relevant and desired features from the data. Thirdly, the feature engineering phase’s output is fed to the training phase to empower the forecaster through training and learning processes to accurately forecast electrical energy consumption. Finally, the forecasted electrical energy consumption is given as an input to the optimization phase to further minimize the error in predicted results by optimizing the model’s hyperparameters. The first stage results are fed to the second stage of the proposed model for efficient energy management. The second stage is based on optimization strategies that utilize the forecasted electrical energy consumption pattern for efficient energy management. The second stage comprises three modules: (i) day ahead genetic modified enhanced differential evolution algorithm based module, (ii) genetic modified enhanced differential evolution algorithm based scheduling module, and (iii) genetic wind driven optimization algorithm based energy management controlling module. The purpose of the second stage is to reduce the bill of electricity, mitigate peaks in demand, and acquire the desired tradeoff between electricity bill and user discomfort by utilizing forecasted electrical energy consumption. The proposed model is favorable for both consumers and power companies because it fulfills the need both parties. For consumers, the proposed model minimizes electricity bill and discomfort in terms of waiting time simultaneously. In contrast, the proposed model rewards power companies by alleviating peaks in demand to increase power system stability. Simulation results confirmed the effectiveness and productiveness of the proposed model by comparing it with benchmark models.