AI-Based Forecasting: Parameters Summary

Forecasting Parameters

Parameter TypeDetails
Forecast Horizons15 min to 7 days
Measurement Frequency10 seconds
Data DurationJan 2020 – Nov 2024
Total Data Points17,297,280

Input Features

  • Group 1: First derivatives – 𝑑GHI, 𝑑GHIcs, 𝑑kcs
  • Group 2: Second derivatives – 𝑑²GHI, 𝑑²GHIcs, 𝑑²kcs
  • Group 3: Third derivatives – 𝑑³GHI, 𝑑³GHIcs, 𝑑³kcs
  • Group 4: Basic – GHI, GHIcs, kcs
  • Group 5: Seasonal – hour, day, month (sin/cos encoding)

Performance Evaluation Metrics

MetricPurpose
MAEMean Absolute Error
RMSERoot Mean Square Error
nMAENormalized MAE
nRMSENormalized RMSE
Coefficient of Determination
95% CIConfidence Interval

Model Architecture

  • Model: LSTM (Long Short-Term Memory)
  • Layers: 3 LSTM layers with 300 units each
  • Final Layer: Dense (1 unit)
  • Activations: Sigmoid, Tanh
  • Training: Early stopping, time-series CV
  • Framework: Keras (Python)

Data Preprocessing Steps

  1. Missing value handling
  2. Outlier detection
  3. Clear-sky irradiance with PVLib
  4. Remove night-time data
  5. Compute kcs (clear sky index)
  6. Compute 1st–3rd derivatives
  7. Generate seasonal features
  8. Downsampling for multiple horizons

Feature Engineering Settings

  • Polynomial Order: Varies (1–5) based on horizon
  • History Window (h): 5 or 10 time steps
  • Feature Selection: Grid search based

✅ Advantages

AspectDetails
High AccuracyLSTM-based model achieved low MAE and RMSE.
Long-term Dataset6 years of high-frequency data (17M+ samples).
Comprehensive FeaturesDerivatives up to 3rd order, seasonality, clear sky indices.
Multi-Horizon ForecastingFrom 15-minute to 7-day forecasts.
Open Dataset & CodeShared via GitHub for reproducibility.
Preprocessing PipelineIncludes normalization and feature engineering.

⚠️ Limitations

LimitationExplanation
High Computational DemandTraining requires substantial hardware.
Data-DependentPerformance may drop outside original region.
Limited ExplainabilityDeep LSTMs are black-box models.
No Real-time IntegrationNo deployment for MPPT/load shifting.
Lack of Ensemble ComparisonNo hybrid baselines included.
Scalability ConstraintsArchitecture not designed for microgrids.

🔧 Possible Modifications & Improvements

SuggestionBenefit
Use Hybrid/Ensemble ModelsImprove generalization and interpretability.
Deploy in Real-Time SystemsSmart grid decisions based on forecasts.
Use Transfer LearningBetter generalization in new regions.
Explainable AIUse SHAP/LIME to interpret model.
Edge DeploymentEnable Raspberry Pi/Jetson deployment.
Integrate Weather APIsBoost forecast accuracy in real-time.