A Comparative Analysis of Short Term Load Forecasting Using LSTM, CNN, and Hybrid CNN-LSTM

Short-term load forecasting (STLF) is becoming more significant and useful in the field of power systems. This manuscript proposes the best approach by comparing the results of three different deep learning models LSTM (Long-Short Term Memory), CNN (Convolutional Neural Network), and Hybrid CNN-LSTM for forecasting the short-term load, and each model is trained using sliding window algorithm and analyzed with statistical parameters like Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), R \(^2\) , Mean Absolute Error (MAE), and training time. The models are implemented in the Tensor Flow platform and executed in Google Colab. Results show that Hybrid models give better results as compared to individual models. In this case it is shown that Hybrid CNN-LSTM gave better result as compared to CNN and LSTM individually. The RMSE, R \(^2\) , MSE, MAE, and MAPE of Hybrid CNN-LSTM model are 0.024225, 0.96, 0.00058, 0.01567, and 0.051564 respectively.