Big Data Analytics for Demand Response in Smart Grid

Abstract

The power industry will depend on Smart Grid (SG) to a great degree in the future. It provides qualitative and quantitative services for better management of energy. Electrical devices such as Advanced Metering Infrastructure (AMI) and Smart Meter (SM) produce large data which is called Big Data. These Big Data is in the form of time series data that requires complex data analytics for prediction of consumption of energy. Prediction of consumption of energy using Big Data analytics can help to balance the supply and demand of energy which is one of the challenging task of SG. The researchers have covered these topics however, they have not tuned the parameters with optimization algorithm such as Genetic Algorithm (GA) for time series data. They have not analysed the prediction of energy using the Prophet model, data anomaly detection techniques and filtering techniques with respect to large time series data in SG. In the first scheme, GA is applied for tuning the parameters of Long Short Term Memory (LSTM). GA is an evolutionary process which is used for optimization. LSTM memorises values over arbitrary intervals which are capable to manage time series data. GA is combined with LSTM in order to process hyper-parameters such as hidden layers, epochs, data intervals, batch size and activation functions. Hence, GA creates a new vector for optimum solution that provides minimum error. These methods provide better results when compared with existing benchmarks. Moreover, GA-LSTM is used in a multi-threaded environment which will increase the speed of convergence. In the second scheme, various filtering techniques are used to predict the energy forecast which can improve the quality of service to the users. The filtering techniques ’ primary task is to handle non-linearity in the input dataset. Various filtering techniques reduce the redundant data for energy consumption prediction. Five different filtering techniques such as Butterworth, Smoothing, Kalman, Frequency, and Filtfilt have been used to preprocess the five different power consumption datasets. LSTM model was used on the processed data for the power consumption prediction. In third scheme, Auto Regressive Integrated Moving Average (ARIMA) and Prophet model is used for energy prediction. ARIMA is mainly used by professionals who have prior knowledge of the intricacies of the model. If a single parameter in the equation is incorrect, the entire result will be affected. However, the Prophet model uses a Bayesian curve fitting method and does not require prior knowledge of datasets. It automatically finds seasonal trends from the data. The Prophet model incorporates seasonal trends such as holidays and weekends, whereas the ARIMA model incorporates both seasonal and non-seasonal trends with time-series data. It provides great precision compared to any other method. The fourth scheme, uses anomaly detection techniques for the large datasets. Different anomaly techniques are compared and tested as preprocessing techniques with LSTM, ARIMA and Prophet models and the results are analysed with different performance metrics. Different anomaly techniques such as forest, K-NN, Histogram, SVM, SOS, and OSVM have been used and compared with preprocessing algorithm on different datasets.The novelty of the work lies in the preprocessing techniques on the LSTM, ARIMA and Prophet model where different anomaly techniques have been compared.