Hybrid Cnn-Lstm Model with Multivariate Data to Increase the Forecast Accuracy of Electricity Consumption

The higher-than-ever demand for electricity strains global power supply systems. Accurate forecasting of energy demand and consumption is required globally to maintain a nation’s lifestyle and economic standards. However, climate change has been impeding efforts to accurately predict the energy demands of local, national, and global power grids. Moreover, the COVID-19 outbreak changed energy consumption patterns, necessitating more accurate forecasting. Studies have suggested that LSTM and CNN models can curve the peculiar nature of the electric demand. Models were trained based on either electricity loads and weather observations or the national figures of a population, such as GDP, imports, and exports. This segregation results in an inferior forecasting performance. To improve this, we introduce a CNN-LSTM model using a multivariable augmentation approach. Based on previous studies, we use 1D convolution and pooling as an effective method for extracting undiscovered features from a temporal sequence. LSTM outperforms RNN on vanishing gradient problems and retains the benefits of RNN on time-series variables. The proposed model achieves a near-perfect forecasting performance for electricity consumption, surpassing other architectures. State-level analysis and training demonstrated the utility of the proposed methodology in forecasting regional energy consumption, outperforming other models in most areas.