Articles

Electrical Load Forecasting in Developing Regions: Khost Province Considering Socio-Economic and Environmental Factors

/0 Comments/in /by

Accurate long-term electricity load forecasting is essential for strategic power system planning but remains challenging in developing regions like Khost, Afghanistan, due to frequent load shedding and data limitations. This study develops a comprehensive forecasting framework using a Multivariate Long Short-Term Memory (LSTM) model. To mitigate data distortions caused by supply constraints, a Standard Load Profile (SLP) was constructed to represent the underlying electricity demand accurately. The model integrates historical consumption with five key explanatory variables: population growth, Gross Domestic Product GDP development, ambient temperature, humidity, and public holidays. Correlation analysis identified temperature as the primary driver of demand, followed by economic growth. The optimized LSTM architecture, featuring two layers (128 and 64 neurons), achieved a Mean Absolute Percentage Error (MAPE) of 3.67%, demonstrating high predictive accuracy. Furthermore, demand was forecasted for the period 2026–2030 under Base, High-Growth, and Low-Growth scenarios. Monte Carlo simulations were employed to generate probabilistic forecasts (P50, P90, and P95), providing a robust quantification of future uncertainty. The findings indicate a consistent rise in future demand, offering a reliable scientific basis for sustainable energy policy and risk-informed infrastructure investment in Khost Province.

Hybrid Bootstrap–LSTM Model for Probabilistic Sea Level Rise Prediction

/in /by

Sea level rise poses increasing risks to coastal regions, highlighting the need for accurate and reliable forecasting methods. This study proposes a probabilistic sea level forecasting framework by integrating a Long Short-Term Memory (LSTM) model with the Moving Block Bootstrap (MBB) technique. The LSTM model is used to capture nonlinear temporal dependencies in sea level time-series data, while the bootstrap approach is employed to quantify prediction uncertainty through probabilistic forecasting. The LSTM model achieved high deterministic prediction accuracy with an MSE of 2.11 × 10!”, RMSE of 0.00459, MAE of 0.00356, and MAPE of 0.34%. The proposed hybrid MBB–LSTM model generates probabilistic forecasts with a 95% confidence interval, resulting in an MSE of 0.01155, RMSE of 0.10749, MAE of 0.08370, and MAPE of 8.99%. Forecast results indicate relatively stable sea level variability until 2026 with an estimated rising trend of approximately 7.44 mm per year. The proposed hybrid framework provides a more informative prediction approach by combining deep learning with bootstrap-based uncertainty estimation, which is valuable for coastal risk assessment and climate adaptation planning.