A hybrid model for building energy consumption forecasting using long short term memory networks

Data driven building energy consumption forecasting models play a significant role in enhancing the energy efficiency of the buildings through building energy management, energy operations, and control strategies. The multi-source and heterogeneous energy consumption data necessitates the integration of evolutionary algorithms and data-driven models for better forecast accuracy and robustness. We present eDemand, an energy consumption forecasting model which employs long short term memory networks and improved sine cosine optimization algorithm for accurate and robust building energy consumption forecasting. A novel Haar wavelet based mutation operator was introduced to enhance the divergence nature of sine cosine optimization algorithm towards the global optimal solution. Further, the hyperparameters (learning rate, weight decay, momentum, and number of hidden layers) of the LSTM were optimized using the improved sine cosine optimization algorithm. A case study on the real-time energy consumption data obtained from Kanwal Rekhi building, an academic building at Indian Institute of Technology, Bombay for short, mid, and long-term forecasting. Experiments reveal that the proposed model outperforms the state-of-the-art energy consumption forecast models in terms of mean absolute error, mean absolute percentage error, mean square error, root mean square error, and Theil statistics. It is shown that stable and accurate forecast results are produced by ISCOA-LSTM and hence it can be used as an efficient tool for solving energy consumption forecast problems.