SolarBoost: Distributed Photovoltaic Power Forecasting

SolarBoost is an advanced boosting method for distributed photovoltaic (DPV) power forecasting, developed by DAMO Academy, Alibaba Group. This repository contains the implementation and experimental code for the SolarBoost algorithm.

Features

• Accurate forecasting for distributed photovoltaic power systems
• Boosting-based methodology for improved prediction accuracy
• Support for multiple datasets including AR, Kalman, and real-world city data
• Comprehensive experimental analysis and benchmarking

Quick Start Guide

To reproduce our experimental results, run exp.py with one of the following options:

1. python exp.py table2 - Reproduces Table 2 (AR grid analysis)
2. python exp.py table3 - Reproduces Table 3 (Aggregate output analysis for AR and Kalman datasets)
3. python exp.py table4 - Reproduces Table 4 (City A dataset analysis)
4. python exp.py figure9 - Generates capacity plots for Figure 9

Project Structure

• ./data/ - Contains the datasets
  • 0.2.npz - Real-world data from City A
• ./models/ - Stores trained models
• ./figures/ - Output directory for generated figures
• ./tables/ - Output directory for result tables

Model Training

Models are automatically trained if they don't exist for the specified parameters. Configuration parameters can be found in:

• ar1.py - AR model parameters
• kalman.py - Kalman filter parameters
• city_a.py - City A dataset parameters