Deep autoencoder with localized stochastic sensitivity for short-term load forecasting

• Training error as the objective function causes overfitting for autoencoders. A short-term electric load forecasting model with D-LiSSA is proposed. performance evaluated using four real-world public electricity datasets. Increased accuracy by 63.20% compared to state-of-the-art models. This paper presents a based on deep autoencoder localized stochastic sensitivity (D-LiSSA). can learn informative hidden representations from unseen samples minimizing perturbed (including training and sensitivity) historical data. Specifically, this general network learning improves prediction reliability. Moreover, nonlinear fully connected feedforward neural regression layer applied forecast load, generalization capability of proposed learned D-LiSSA. The markets France (FR), Germany (GR), Romania (RO), Spain (ES) ENTSO-E. Extensive experimental results comparisons classical models show that yields accurate achieves desired reliable capability. For instance, French case, lowest mean absolute error, percentage root squared error; providing up 61.89%, 63.20%, 56.40% improvements benchmark hourly horizon, respectively.