TL;DR
This paper introduces a two-stage stochastic optimization framework for day-ahead operational planning in urban microgrids with significant photovoltaic (PV) integration. It addresses the challenge of PV production uncertainty by optimizing generator scheduling and reserve allocation to minimize both economic costs and carbon emissions, achieving significant savings compared to traditional deterministic approaches.
What Problem Does This Paper Address?
The increasing integration of renewable energy sources (RES), particularly solar photovoltaics (PV), into power grids presents a significant challenge due to their inherent variability and unpredictability. This uncertainty makes traditional day-ahead operational planning, which relies on deterministic forecasts, insufficient to guarantee grid stability and economic efficiency. Specifically, the paper addresses:
- PV Production Uncertainty: PV output fluctuates due to weather conditions, making accurate forecasting difficult. This variability can lead to power imbalances.
- Operating Reserve (OR) Quantification: Determining the appropriate amount of operating reserve needed to compensate for these uncertainties is crucial for grid security but challenging to quantify precisely. Over-scheduling OR leads to unnecessary costs, while under-scheduling risks power deficits.
- Economic and Environmental Costs: The need to balance the grid often requires the dispatch of fast-start conventional generators, which can increase fuel costs and carbon emissions, especially during rapid ramp-ups.
- Urban Microgrid Specifics: The paper focuses on urban microgrids, which have unique characteristics like distributed energy resources and potentially limited inertia, requiring tailored optimization approaches.
The authors aim to develop a robust operational planning strategy that can effectively manage PV uncertainty, optimize reserve allocation, and minimize both economic and environmental impacts.
Methods
The paper proposes a novel two-stage stochastic optimization framework to tackle the operational planning problem. This approach breaks down the decision-making process into two distinct stages:
- Stage 1: Deterministic Operational Planning with Risk Assessment:
- Reserve Quantification: A risk-constrained probabilistic method is used to determine the required operating reserve (OR). This involves analyzing historical PV and load forecasting errors to build probability density functions (PDFs) and then using a Loss of Load Probability (LOLP) index to set a desired security level (risk index, ε).
- Deterministic Unit Commitment (DUC): Based on the forecasted net demand and the calculated OR, a Mixed-Integer Linear Programming (MILP) model is employed to determine the optimal commitment and dispatch of slow-start conventional generators for the next day. This stage aims to minimize operating costs and/or CO2 emissions.
- Stage 2: Stochastic Operational Planning:
- Scenario Generation: Future uncertainties in PV production are represented by multiple scenarios derived from the probability distributions of forecasting errors. This allows for a more nuanced representation of potential deviations.
- Stochastic Unit Commitment (SUC): The second stage re-optimizes the operational plan considering these scenarios. The commitment of slow generators from Stage 1 is kept fixed, but the dispatch of fast-start generators (including PV active generators) is adjusted to compensate for the specific uncertainties within each scenario. This stage aims to minimize the expected costs (including recourse costs from Stage 1) across all scenarios.
- PV Active Generators (PV AGs): PV generators are treated as "PV Active Generators" (PV AGs) that can be curtailed to provide operating reserve, offering a flexible and cost-effective reserve source.
The optimization objectives can be mono-objective (minimizing cost or emissions) or multi-objective (minimizing both simultaneously).
Datasets
The paper utilizes historical data for PV power and load demand forecasting to generate forecast errors. A homemade Back-propagation Neural Network (BPNN) is mentioned for generating PV power and load forecast data with confidence intervals. The specific datasets used for training and evaluation are not explicitly detailed beyond their origin from a local energy community in Lille, France. No new benchmark datasets are introduced.
Key Insights and Ideas
- Probabilistic Reserve is Superior to N-1: The study demonstrates that traditional N-1 criteria for reserve scheduling are often overly conservative and inefficient, leading to over-scheduled reserves. Probabilistic methods, especially the scenario-based stochastic approach, provide a more accurate and efficient way to quantify and allocate reserves.
- PV Curtailment as a Reserve Source: Treating PV generation as a controllable reserve source (PV AGs) by allowing for planned curtailment significantly enhances the flexibility and cost-effectiveness of reserve provision.
- Two-Stage Optimization Captures Uncertainty Better: The two-stage approach effectively bridges the gap between deterministic planning and the reality of uncertain future events. Stage 1 provides a robust baseline, while Stage 2 refines the plan to adapt to specific uncertain outcomes, leading to better cost and emission performance.
- Significant Cost and Emission Reductions: The stochastic optimization approach, particularly the second stage, leads to substantial savings in operating costs and carbon emissions compared to deterministic methods, highlighting the economic and environmental benefits of accounting for uncertainty.
- Impact of RES Penetration: The study shows that as the PV self-production rate increases, the benefits of the stochastic optimization approach become more pronounced, with larger differences in cost and emissions compared to deterministic methods.
Key Results
- Cost and Emission Savings: The proposed scenario-based stochastic optimization approach (2nd stage) achieved a reduction of approximately 15% in economic operating costs and environmental costs compared to the deterministic N-1 criterion.
- Improved Reserve Utilization: The stochastic optimization resulted in a more reasonable and higher reserve utilization rate, with the daily reserve energy from fast generators increasing by 15% compared to the deterministic case.
- Higher Security Level: The 2nd stage optimization provides a higher level of power security due to more effective reserve power availability, ensuring that the reserve requirement is met even in worst-case scenarios.
- PV Self-Consumption: The study demonstrated an increase in PV self-consumption rate from 17% (N-1 criterion) to 67% (deterministic optimization with risk assessment) and then to a higher level with the stochastic approach, indicating better utilization of locally generated renewable energy.
Limitations and Future Work
- Limitations:
- The paper acknowledges that the optimization application uses simplified linear cost functions for generators.
- The study assumes that voltage drops and line losses are negligible in the urban microgrid, which might not hold true for larger or more complex networks.
- The scenario generation is based on normal distribution approximations, which might not capture all types of forecasting error distributions.
- Future Work:
- The authors suggest integrating storage systems as an alternative for fast power reserve provision, potentially replacing fast generators and further aiding CO2 abatement.
- Further research could explore more complex generator cost models and consider transmission constraints for larger networks.
- Investigating other probability distributions for scenario generation could enhance the robustness of the approach.
Summary updated