Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SolarBoost: Distributed Photovoltaic Power Forecasting Amid Time-varying Grid Capacity

Linyuan Geng, Linxiao Yang, Xinyue Gu, Liang Sun

TL;DR

SolarBoost tackles DPV power forecasting under time-varying grid capacity by decomposing the region into grids and learning a per-unit-capacity output function together with a dynamic generalized-capacity for each grid. The method alternates boosting-based updates of the unit-output model and Kalman-filter-based updates of the grid capacities, guided by a surrogate upper-bound loss that reduces cross-grid entanglement. Theoretical analysis demonstrates grid-level modeling can yield tighter predictive bounds than aggregate approaches, and experiments on synthetic and five-city real-world data show substantial improvements in unit-output, generalized-capacity, and aggregated power forecasts, with deployment evidence highlighting operational and economic benefits. This work presents a practical, scalable approach to integrating capacity dynamics into DPV forecasting, with strong implications for grid planning and renewable integration.

Abstract

This paper presents SolarBoost, a novel approach for forecasting power output in distributed photovoltaic (DPV) systems. While existing centralized photovoltaic (CPV) methods are able to precisely model output dependencies due to uniformity, it is difficult to apply such techniques to DPV systems, as DPVs face challenges such as missing grid-level data, temporal shifts in installed capacity, geographic variability, and panel diversity. SolarBoost overcomes these challenges by modeling aggregated power output as a composite of output from small grids, where each grid output is modeled using a unit output function multiplied by its capacity. This approach decouples the homogeneous unit output function from dynamic capacity for accurate prediction. Efficient algorithms over an upper-bound approximation are proposed to overcome computational bottlenecks in loss functions. We demonstrate the superiority of grid-level modeling via theoretical analysis and experiments. SolarBoost has been validated through deployment across various cities in China, significantly reducing potential losses and provides valuable insights for the operation of power grids. The code for this work is available at https://github.com/DAMO-DI-ML/SolarBoost.

SolarBoost: Distributed Photovoltaic Power Forecasting Amid Time-varying Grid Capacity

TL;DR

SolarBoost tackles DPV power forecasting under time-varying grid capacity by decomposing the region into grids and learning a per-unit-capacity output function together with a dynamic generalized-capacity for each grid. The method alternates boosting-based updates of the unit-output model and Kalman-filter-based updates of the grid capacities, guided by a surrogate upper-bound loss that reduces cross-grid entanglement. Theoretical analysis demonstrates grid-level modeling can yield tighter predictive bounds than aggregate approaches, and experiments on synthetic and five-city real-world data show substantial improvements in unit-output, generalized-capacity, and aggregated power forecasts, with deployment evidence highlighting operational and economic benefits. This work presents a practical, scalable approach to integrating capacity dynamics into DPV forecasting, with strong implications for grid planning and renewable integration.

Abstract

This paper presents SolarBoost, a novel approach for forecasting power output in distributed photovoltaic (DPV) systems. While existing centralized photovoltaic (CPV) methods are able to precisely model output dependencies due to uniformity, it is difficult to apply such techniques to DPV systems, as DPVs face challenges such as missing grid-level data, temporal shifts in installed capacity, geographic variability, and panel diversity. SolarBoost overcomes these challenges by modeling aggregated power output as a composite of output from small grids, where each grid output is modeled using a unit output function multiplied by its capacity. This approach decouples the homogeneous unit output function from dynamic capacity for accurate prediction. Efficient algorithms over an upper-bound approximation are proposed to overcome computational bottlenecks in loss functions. We demonstrate the superiority of grid-level modeling via theoretical analysis and experiments. SolarBoost has been validated through deployment across various cities in China, significantly reducing potential losses and provides valuable insights for the operation of power grids. The code for this work is available at https://github.com/DAMO-DI-ML/SolarBoost.
Paper Structure (38 sections, 3 theorems, 23 equations, 9 figures, 5 tables, 1 algorithm)

This paper contains 38 sections, 3 theorems, 23 equations, 9 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. DPV Power Forecasting as a Regression Task
  4. Impact of Changing Capacity
  5. Limitations of Ignoring Grid Capacity
  6. Problem Formulation
  7. SolarBoost
  8. Surrogate Loss Function
  9. Update Unit Output Function $f$ via Boosting
  10. Update Generalized-Capacity $c_{i,t}$
  11. Overall Framework
  12. Theoretical Analysis
  13. Experiment
  14. Setup
  15. Baselines
  16. ...and 23 more sections

Key Result

lemma 1

A convex function $\kappa(\cdot)$ satisfies the inequality: where $\alpha_i \geq 0$ and $\sum_i \alpha_i = 1$ for any collection of points $t_i$ and corresponding multipliers $\alpha_i$. For a linear function $\bm{z}^T \bm{\beta}$, the following inequality holds:

Figures (9)

  • Figure 1: Relationship between installed capacity and power output. Dark orange curve: actual DPV power output in an eastern Chinese city. Light orange shade: the installed capacity of this city. Dashed blue line: initial capacity for reference. Capacity expansion lifts the potential output ceiling, with daily output fluctuations governed by weather.
  • Figure 2: Varying grid-level capacities proportions (b) across time make grid-level modeling necessary: relying solely on total capacity (orange curve in (c)) results in large errors (difference between orange and blue dashed lines in (c)), even under similar weather conditions between day 2 and 3 in (a).
  • Figure 3: The framework of SolarBoost, which uses a surrogate loss function to approximate the true objective function and alternating updates of unit output predictor and generalized-capacity by boosting and Kalman filter, respectively.
  • Figure 4: From top to bottom, the cases are for sunny and overcast days. From left to right: (a) Line graph comparing actual power output with model predictions. (b) Heatmap of irradiation at 12 PM. (c) Heatmap of cloud cover at 12 PM.
  • Figure 5: RMSE vs. number of grids (a) and lambda (b) across five cities (A-E).
  • ...and 4 more figures

Theorems & Definitions (3)

  • lemma 1
  • theorem 1: Superiority of Grid-Level Modeling
  • theorem 2