[Paper Review] Intelligent Subset Selection of Power Generators for Economic Dispatch

Biswarup Bhattacharya, Abhishek Sinha|arXiv (Cornell University)|Sep 8, 2017

Energy Load and Power Forecasting10 references3 citations

TL;DR

This paper proposes a deep learning-based approach for intelligent subset selection of solar power generators in economic dispatch, using a 23-dimensional input (20 bus voltages + 3 generator predictions) to minimize a composite loss of prediction error and network congestion. The model achieves low test losses (L1: 8, L2: 59), demonstrating effective, scalable optimization for renewable-rich power grids.

ABSTRACT

Sustainable and economical generation of electrical power is an essential and mandatory component of infrastructure in today's world. Optimal generation (generator subset selection) of power requires a careful evaluation of various factors like type of source, generation, transmission & storage capacities, congestion among others which makes this a difficult task. We created a grid to simulate various conditions including stimuli like generator supply, weather and load demand using Siemens PSS/E software and this data is trained using deep learning methods and subsequently tested. The results are highly encouraging. As per our knowledge, this is the first paper to propose a working and scalable deep learning model for this problem.

Motivation & Objective

• To address the challenge of optimal generator subset selection in power systems under dynamic load and weather conditions.
• To reduce economic losses from power prediction errors and network congestion in renewable energy integration.
• To develop a scalable, data-driven method for real-time economic dispatch using deep learning.
• To enable smarter, more efficient operation of renewable-rich power grids through intelligent generator selection.

Proposed method

• A deep neural network with one hidden layer of 200 ReLU-activated neurons processes a 23-dimensional input vector combining bus voltage magnitudes and predicted solar power for three generators.
• The model minimizes a composite loss function 𝒟 = (model output − (L₁ + L₂))², where L₁ is prediction error and L₂ is a binary congestion penalty (50 if congested, 0 otherwise).
• Training data (4,500/4,900 points) was generated via PSS/E simulations of a 20-bus network with three solar generators under varying weather and load conditions.
• The model uses the Adam optimizer to minimize the loss, with L₁ scaled to balance magnitude with the binary L₂ penalty.
• All subsets of the three solar generators were evaluated, and the model learned to predict total cost (error + congestion) for each subset.
• Test performance was evaluated on 400 held-out simulation points to assess generalization.

Experimental results

Research questions

• RQ1 Can a deep learning model effectively predict the optimal subset of solar generators to minimize economic loss in power dispatch?
• RQ2 How well can a neural network generalize to unseen load and weather conditions in subset selection for economic dispatch?
• RQ3 To what extent does incorporating congestion penalties improve the economic and operational reliability of generator subset selection?
• RQ4 Can a machine learning model outperform traditional optimization methods in dynamic, renewable-integrated grid scenarios?

Key findings

• The model achieved a test L₁ loss of 8, indicating strong generalization in predicting actual power output for selected generator subsets.
• The test L₂ loss was 59, reflecting the model’s ability to identify subsets that avoid network congestion with high accuracy.
• The training and test loss curves showed stable convergence, with minimal overfitting, suggesting robustness to unseen data.
• The composite loss function effectively balanced prediction accuracy and congestion avoidance, enabling economically optimal subset selection.
• The model demonstrated scalability and practicality for deployment in real-time economic dispatch systems with renewable sources.
• The results validate the feasibility of using deep learning for intelligent generator subset selection in smart grids, particularly under variable solar generation.