[Paper Review] The Concept of the Deep Learning-Based System "Artificial Dispatcher" to Power System Control and Dispatch

Nikita Tomin, Victor Kurbatsky+1 more|arXiv (Cornell University)|May 7, 2018

Industrial Engineering and Technologies5 references3 citations

TL;DR

This paper proposes the 'Artificial Dispatcher,' a deep learning-based intelligent system that integrates human dispatcher expertise with the speed of automated control to enhance power system stability and emergency response. By leveraging deep neural networks (DNNs) and convolutional neural networks (CNNs), the system enables real-time voltage stability assessment and optimal reactive power control, reducing error rates and computation time from minutes to seconds while improving accuracy under data corruption.

ABSTRACT

Year by year control of normal and emergency conditions of up-to-date power systems becomes an increasingly complicated problem. With the increasing complexity the existing control system of power system conditions which includes operative actions of the dispatcher and work of special automatic devices proves to be insufficiently effective more and more frequently, which raises risks of dangerous and emergency conditions in power systems. The paper is aimed at compensating for the shortcomings of man (a cognitive barrier, exposure to stresses and so on) and automatic devices by combining their strong points, i.e. the dispatcher's intelligence and the speed of automatic devices by virtue of development of the intelligent system "Artificial dispatcher" on the basis of deep machine learning technology. For realization of the system "Artificial dispatcher" in addition to deep learning it is planned to attract the game theory approaches to formalize work of the up-to-date power system as a game problem. The "gain" for "Artificial dispatcher" will consist in bringing in a power system in the normal steady-state or post-emergency conditions by means of the required control actions.

Motivation & Objective

• Address the growing complexity and instability in modern power systems due to increasing disturbances and limited effectiveness of current dispatch and emergency control systems.
• Overcome human cognitive limitations and stress-induced errors in dispatcher decision-making under time pressure and uncertainty.
• Enhance the reliability and survivability of power systems by integrating intelligent decision-making with existing SCADA/EMS/DMS infrastructure.
• Develop a system that can autonomously simulate dispatcher actions in real time, reducing reliance on human operators during normal and emergency conditions.
• Improve the efficiency and speed of power system control through machine learning models capable of processing large-scale data and identifying optimal control actions rapidly.

Proposed method

• Utilize deep neural networks (DNNs) and convolutional neural networks (CNNs) to model and predict power system behavior, particularly focusing on voltage stability.
• Implement online decision trees and real-time learning algorithms to dynamically assess the L-index (voltage stability indicator) using live SCADA data.
• Train the system on historical and simulated power system data to learn optimal reactive power injection strategies for emergency and normal conditions.
• Integrate the 'Artificial Dispatcher' as an additional module within existing SCADA/EMS/DMS platforms to leverage their data collection, analysis, and control functions.
• Apply game theory principles to formalize system control as a strategic decision-making problem, where the goal is to restore or maintain system stability.
• Use elastic weight consolidation and Boltzmann machine-based learning techniques to enable continual learning and memory retention across diverse system states.

Experimental results

Research questions

• RQ1 Can deep learning models effectively simulate dispatcher decision-making in real time under high-stress, time-constrained conditions?
• RQ2 How does the performance of a deep learning-based system compare to traditional algorithmic methods in computing voltage stability indices under data corruption?
• RQ3 To what extent can a DNN/CNN-based system reduce computation time for optimal control actions while maintaining or improving accuracy?
• RQ4 Can the system reliably identify and apply corrective control actions (e.g., reactive power injections) to prevent voltage collapse in complex power systems?
• RQ5 How effectively can the system integrate with existing SCADA/EMS/DMS platforms to enhance, rather than replace, current control infrastructure?

Key findings

• The deep learning model reduced root-mean-square error in L-index estimation to approximately 13% when tested on corrupted IEEE 118-bus system data, outperforming traditional methods.
• Computation time for determining optimal reactive power injections was reduced from 30–40 minutes using conventional methods to just a few hundred milliseconds (centiseconds) using the machine learning approach.
• The system demonstrated robustness to 'bad data' and missing measurements, maintaining accurate stability assessments even when input data was corrupted.
• After applying corrective control actions predicted by the model, the sum of local voltage stability indices ($L_{sum}$) decreased significantly, indicating reduced risk of voltage collapse.
• The system's ability to process large-scale data and identify optimal control strategies in fractions of a second exceeds the physiological limits of human dispatchers.
• Integration of the 'Artificial Dispatcher' as a supplementary module to existing SCADA/EMS/DMS systems enables real-time, high-speed, and accurate control without disrupting current operational workflows.