Electricity Load Forecasting using Hybrid SSA–Deep Learning

Objective

Developed an optimal Hybrid SSA–Deep Learning model by integrating SSA with LSTM, BiLSTM, and GRU architectures to improve short-term electricity load forecasting accuracy.

Project Details

Developed an end-to-end forecasting pipeline for hourly electricity load (~77,000 data points) using Hybrid SSA + LSTM/BiLSTM/GRU models built with TensorFlow and Scikit-learn.
Implemented SSA decomposition for feature engineering, effectively separating noise, trend, and seasonal components (hourly, daily, weekly), resulting in cleaner and more accurate model inputs.
Conducted comprehensive hyperparameter optimization, achieving prediction accuracy > 95% with RMSE ±80 MW.
Performed statistical analysis and time series visualization to examine trends, seasonality, and fluctuations, providing valuable insights for PLN’s capacity planning.

Key Results

✓**Achieved >95% prediction accuracy** in electricity load forecasting using a Hybrid SSA–Deep Learning model.
✓Processed and analyzed **~77,000 hourly electricity load data points** to generate forecasts supporting power system capacity planning.

Comments (1)

Rian HidayatJul 21, 2026

Penerapan dekomposisi SSA sebelum LSTM berhasil menaikkan akurasi di atas 95%. Riset dan karya tesis yang sangat solid!

Project Information

Date:Oct 2025

Duration:6 months

Tools & Technologies

TensorFlowScikit-learnPythonPandasNumPyMatplotlibLSTMBiLSTMGRUSingular Spectrum Analysis (SSA)Time Series ForecastingHyperparameter OptimizationData AnalysisStatistical AnalysisData Visualization