A Data-Driven Hybrid Optimization Based Deep Network Model for Short-Term Residential Load Forecasting

Over the last few decades, electricity consumption has been increasing at an exponential rate. The power distributors are being put under a lot of strain as a result of such rise. The estimation of energy usage at the consumer level is a possible solution to this problem. In this paper, we offered various Deep Learning (DL) and hybrid DL optimization models, particularly Convolutional Neural Network (CNN), Long-Short-Term Memory (LSTM), CNN-LSTM, and ConvLSTM, to predict short-term electric charge consumption for residents to completely receive the benefits of time series mining technologies. Extensive experimental findings indicate that the hybrid optimization technique may greatly enhance the precision and resilience of short-term residential charge predictions. Both root mean squared error (RSME) and mean squared error (MSE) has been used to assess the model's performance. Research results demonstrate that hybrid network (CNN-LSTM) outperforms the other DL models and the state of art models with an RSME of 0.055 and MSE of 0.0030 on test data.