Overview
This experimental study investigates the application of Scenario-based Data-enabled Predictive Control (Scenario-DeePC) to a Battery Energy Storage System (BESS). The core objective addresses the challenge of maintaining strict state-of-charge (SOC) limits within BESS operations, which is complicated by factors such as measurement noise and inherently nonlinear dynamics that are difficult to model. The research focuses on how Scenario-DeePC, an extension of Data-enabled Predictive Control (DeePC), manages constraint robustness through a data-driven approach, particularly in a real-world setting.
Research Context
Battery energy storage systems require strict adherence to state-of-charge (SOC) limits to prevent operational issues such as overcharge and deep discharge. Achieving this control is complex due to the presence of measurement noise and the nonlinear dynamics of batteries, which are often costly to model. Traditional Data-enabled Predictive Control (DeePC) addresses the modeling challenge by predicting future behavior directly from observed data. However, the regularization inherent in standard DeePC primarily enhances prediction robustness, not necessarily constraint robustness. This limitation motivated the development and testing of Scenario-DeePC, which aims to build constraint robustness directly from observed prediction errors rather than relying on assumed disturbance distributions.
Approach
The study deployed Scenario-DeePC on a grid-connected battery system located at the NEST research facility. This deployment represents the first real-world implementation of the Scenario-DeePC methodology. The system's state-of-charge (SOC) estimate was noted to exhibit abrupt and irregular recalibration jumps, in addition to typical measurement noise. The core mechanism of Scenario-DeePC involves extending DeePC with the scenario approach, which enables data-driven construction of constraint robustness based on observed prediction errors. This contrasts with methods that presuppose a particular disturbance distribution.
Findings
- Scenario-DeePC achieved tracking performance comparable to that of standard DeePC.
- Scenario-DeePC exhibited substantially fewer constraint violations compared to standard DeePC, particularly concerning the maintenance of strict SOC limits.
- The adaptive scenario buffer within Scenario-DeePC automatically tightened constraint handling.
- This adaptive buffer improved robustness to unpredictable recalibration events observed in the battery system's SOC estimate.
Why This Matters
The findings indicate a method for more robust control of battery energy storage systems, crucial for preventing overcharge and deep discharge. The data-driven approach of Scenario-DeePC offers a way to manage complex, costly-to-model nonlinear battery dynamics and unexpected measurement anomalies like SOC recalibration jumps. Its ability to maintain comparable tracking performance with fewer constraint violations suggests potential for improved operational reliability of BESS in real-world grid applications.
Potential Applications
The successful real-world deployment of Scenario-DeePC on a grid-connected battery system suggests its applicability for managing battery energy storage systems where strict state-of-charge limits are critical. Its robustness to measurement noise and irregular recalibration events could make it suitable for diverse battery control scenarios.