TL;DR
This paper introduces a "risk-averse" model predictive control (MPC) strategy for islanded microgrids with a high proportion of renewable energy sources. It tackles the challenge of uncertain renewable energy generation and load by allowing designers to tune the trade-off between performance and safety, making microgrid operation more robust to prediction errors.
What Problem Does This Paper Address?
The increasing integration of renewable energy sources (RES) like solar and wind into power grids presents a significant challenge for grid operation. Unlike conventional power plants, RES generation is inherently intermittent and difficult to predict accurately. This uncertainty, coupled with fluctuating electricity demand, complicates the stable and efficient operation of microgrids (MGs), especially when they are islanded (disconnected from the main grid). Traditional control methods often struggle with these uncertainties, leading to potential constraint violations and suboptimal performance. Existing approaches include:
- Certainty-equivalent MPC: Assumes perfect forecasts, which is unrealistic and can lead to significant issues.
- Worst-case MPC: Overly conservative, as it aims to minimize costs under the absolute worst possible scenario, often sacrificing efficiency.
- Stochastic MPC: Relies on accurate probability distributions of forecasts, which can be hard to obtain and may not fully capture the impact of rare but significant events.
This paper addresses the need for a control strategy that can effectively handle forecast uncertainties and the impact of rare, high-consequence events without being overly conservative.
Methods
The core of the paper is the development of a risk-averse Model Predictive Control (MPC) scheme. Here's a breakdown of the key methodological contributions:
- Risk Measure Formulation: The authors leverage the concept of risk measures, which quantify the significance of extreme events in a probability distribution. They specifically focus on Average Value-at-Risk (AVaR), a coherent risk measure that allows for a tunable level of risk aversion. This is crucial because it bridges the gap between the overly conservative worst-case approach and the potentially brittle stochastic approach.
- Scenario Trees for Uncertainty Representation: To manage the complexity of forecasting, the paper uses scenario trees. These trees provide a compact representation of the probability distribution of future uncertain events (load and RES infeed) over a defined time horizon. This is more efficient than using a large number of independent scenarios.
- MIQCQP Reformulation: The risk-averse MPC problem, which can be complex due to nested risk measures, is reformulated into a Mixed-Integer Quadratically Constrained Quadratic Problem (MIQCQP). This reformulation makes the problem computationally tractable and solvable by standard optimization solvers, enabling real-time applications.
- Tunable Risk Aversion: A key innovation is the ability to tune the risk aversion parameter (α). By adjusting α, the designer can interpolate between a risk-neutral (stochastic) approach (α=1) and a worst-case approach (α=0). This allows for a flexible trade-off between economic performance and operational safety.
Datasets
- Wind Speed and Load Forecasts: The paper utilizes seasonal ARIMA models from a previous work ([20]) to generate forecasts for wind speed and load. These forecasts are then used to create independent scenarios.
- No new benchmark datasets were introduced. The focus is on simulating a microgrid scenario with realistic, albeit generated, data.
Key Insights and Ideas
- Bridging the Gap: Risk measures, particularly AVaR, offer a powerful way to balance the conservatism of worst-case methods with the potential fragility of purely stochastic methods when dealing with uncertain forecasts.
- Tunable Safety: The ability to adjust the risk aversion parameter (α) is a significant practical advantage. It allows grid operators to tailor the control strategy to specific operational needs and risk tolerances, moving beyond a one-size-fits-all approach.
- Robustness to "High-Effect Low-Probability" (HELP) Events: The risk-averse approach is shown to be more resilient to rare but impactful events (like sudden drops in renewable generation or spikes in demand) compared to traditional methods. This is crucial for maintaining grid stability.
- Computational Tractability: The reformulation into an MIQCQP is a critical enabler for applying these advanced control strategies in real-time microgrid operation.
Key Results
- Improved Robustness: The risk-averse MPC demonstrated robustness to forecast errors and uncertainties in probability distributions, as shown through sensitivity analyses.
- Tunable Performance vs. Safety: Simulations showed that by adjusting the risk aversion parameter α, the authors could effectively tune the trade-off between economic costs (lo) and energy-related costs (ls). For instance, a moderate risk aversion (α=0.5) yielded the lowest overall cost in their simulations.
- Reduced Constraint Violations: Compared to a certainty-equivalent approach, the risk-averse method significantly reduced constraint violations, indicating a safer operation.
- Computational Efficiency: The MIQCQP formulation allowed for solve times within acceptable limits for MPC applications, even with a moderate number of scenarios and integer variables.
Limitations and Future Work
- Complexity for Large-Scale Systems: While the MIQCQP reformulation makes the problem tractable, the complexity can still become prohibitive for very large microgrids with many units and a high number of scenario tree nodes.
- Simplified Storage Dynamics: The paper mentions that for the simulation, a more detailed plant model was used than the one directly incorporated into the MPC formulation. Future work could explore more complex storage dynamics within the MPC framework.
- Further Exploration of Integer Variables: The authors plan to investigate methods for handling a larger number of integer variables, potentially through parallel computing or more advanced optimization techniques.
- Chance Constraints and Risk Aversion: Future work will also explore the integration of chance constraints with risk aversion.
- More Complex MG Configurations: The study was performed on a relatively simple MG configuration. Extending the analysis to MGs with more conventional and renewable units is a natural next step.
Summary updated