Source-linked AI summary
The Impact of PV Generation Forecast and Multi-Objective Control Policy on Optimal Operation of Grid Connected PV-BESS Microgrid
Berhane Darsene Dimd, Steve Voller, Ole-Morten Midtgård
TL;DR
Variable PV output makes reliable, efficient operation of grid-connected PV-BESS microgrids dependent on accurate forecasting and effective scheduling. The paper integrates LSTM-based forecasting with MPC-based multi-objective control, finding that LSTM improves operational outcomes over persistence, although with higher battery throughput.
Problem
Variable PV output makes grid-connected PV-BESS microgrid operation dependent on accurate forecasting and effective energy scheduling.
Method
The study integrates LSTM-based PV forecasting with MPC-based multi-objective energy scheduling to balance competing microgrid objectives.
Results
The LSTM approach improved grid-injection reduction, battery utilization, and PV self-consumption over persistence, while perfect forecasts performed best across metrics.
Takeaways & Limitations
Accurate PV forecasting integrated into predictive control is important for efficient microgrid operation and robust handling of forecast uncertainty.
Takeaways & Limitations
The study uses simplified PV and BESS models that lack sufficient complexity to represent real-world dynamics fairly.
Abstract
from arXiv · showhide
The variability of photovoltaic (PV) generation poses significant challenges to the reliable and efficient operation of grid-connected microgrids. Accurate PV output power forecasting and efficient energy scheduling strategies are essential not only for optimizing PV system operation but also for improving the overall performance and reliability of the system. This study proposes a long short-term memory (LSTM)-based PV power forecasting model integrated with a multi-objective scheduling framework for a grid-connected PV-battery energy storage system (BESS). The proposed approach enables detailed performance monitoring and assessment by quantifying how PV forecast accuracy influences key operational metrics, including PV self-consumption ratio, grid energy cost, grid injection, and battery utilization. Three forecasting scenarios (perfect forecast, persistence model, and LSTM-based forecast) are compared to evaluate their impact on system performance and operational reliability. Results show that the LSTM-based forecast reduces root mean squared error (RMSE) by 6% compared with the persistence model, increases the PV self-consumption ratio from 78.1% to 84.5%, and reduces grid injections by 82%. The analysis also highlights trade-offs, as higher battery throughput associated with improved performance may contribute to accelerated aging. These findings demonstrate the importance of accurate PV forecasting in improving system performance and ensuring reliable operation. Future work will focus on probabilistic forecasting to properly quantify uncertainties, incorporate load prediction, and develop smart control strategies that allow grid-support functionalities from the PV side.
1. Introduction
The introduction frames grid-connected PV microgrid operation as a challenge arising from uncertain PV output, battery-aging risks, and grid impacts. It motivates a forecasting-integrated, multi-objective MILP-MPC energy-scheduling framework addressing economic, self-consumption, and grid-integration objectives.
- 30% in 2023 to 37% in 2026: renewable electricity generation is forecast to grow, largely supported by expanding solar PV generation.
- Environmental sensitivity makes grid-connected PV output uncertain and irregular, complicating the balance between variable loads and variable supply.The introduction identifies BESS, PV forecasting, and energy scheduling as tools for microgrid operation.
- Conventional charging of BESS during excess PV increases cycling-related battery degradation and can create large solar-noon grid injections.These injections may contribute to grid congestion, voltage and frequency fluctuations, and infrastructure or curtailment requirements.
- PV forecasting supports dispatch decisions that can avoid unnecessary grid imports and injections, reduce operational costs, and improve renewable-energy utilization.
- The study proposes a holistic MILP- and MPC-based multi-objective energy-scheduling framework balancing cost reduction, local PV consumption, and reduced grid impacts.The framework responds to identified gaps involving simplified system models, insufficient PV-forecast treatment, inadequate grid-impact consideration, and missing real-time grid pricing.
2. Methodology
The methodology uses forecast-informed model predictive control to schedule a grid-connected PV-BESS microgrid while maintaining power balance and optimizing multiple objectives. Battery state of charge, grid exchange, operational constraints, and energy costs are represented in a MILP optimization framework.
- MPC-based energy scheduling: The MPC-based energy scheduler uses PV and consumption forecasts, grid electricity costs, and the current BESS state to determine battery and grid set points.Its goal is to maintain instantaneous microgrid power balance while optimally satisfying multiple control-policy objectives.
- MPC-based energy scheduling: MPC computes future battery, load, and grid control signals over prediction horizon Np while accounting for microgrid dynamics and constraints.At each sampling instant, the first optimized control input is implemented, the remaining inputs are discarded, and the horizon shifts forward one step.
- Microgrid modeling: The system model uses a 15-minute sampling time and represents battery evolution through the Coulomb Counting method for state-of-charge estimation.Battery charging and discharging efficiencies, sampling time, capacity, and battery power are included in the SOC formulation.
- Multi-objective formulation: The objective function combines grid and storage costs, discouraging grid imports and excessive battery cycling while prioritizing PV or battery use.Battery cycling is penalized because excessive cycling can degrade battery life, particularly under variable PV output, load, and spot prices.
- Optimization and evaluation: The resulting MILP optimization problem is formulated and solved in Python using the Gurobi solver over a two-month system-performance analysis.The optimization includes power-balance, battery SOC, and operational constraints.
3. LSTM-based PV Output Forecasting
This section presents an LSTM-based PV generation forecasting model selected for capturing long-term sequential patterns while mitigating vanishing gradients. The model was iteratively optimized on validation data by tuning network hyperparameters to improve forecasting accuracy and test-set generalization.
- Model selection: LSTM was selected because it captures long-term dependencies in sequential PV generation data and uses internal gates to mitigate the vanishing-gradient problem affecting conventional RNNs.The gates help prevent insignificant weight updates associated with vanishing gradients.
- Model optimization: The model was iteratively optimized using validation data by tuning parameters including hidden layers, units per layer, batch size, learning rate, dropout rate, and epochs.The optimization sought a hyperparameter combination that maximized forecasting accuracy and generalization on the test dataset.
- Model optimization: The selected configuration used two hidden layers with 10 units each, a batch size of 24, 150 epochs, and a learning rate of 0.001.These values were identified as optimal and used in the study.
4. Result and Discussion
The results show that the LSTM forecast modestly improves short-term PV prediction over persistence, while forecast quality strongly affects microgrid cost, self-consumption, grid exchanges, and battery utilization. Longer prediction horizons improve SCR and battery throughput up to H=48 before performance declines.
- PV forecast performance: 6% RMSE improvement is achieved by the LSTM-based forecast over the persistence model for 15-minute-ahead prediction.The improvement is modest relative to the LSTM model’s additional complexity and resource requirements.
- Two-month microgrid operation: 90.9% SCR, e37.25 grid cost, and 1182 kWh battery throughput are achieved under perfect forecast conditions.Case 1 imports 4937 kWh and exports 139 kWh, indicating low grid dependence and effective battery utilization.
- Two-month microgrid operation: 78.1% SCR and 419 kWh battery throughput occur with the persistence forecast, alongside 5362 kWh imports and 623 kWh exports.The resulting grid cost is e38.03, only slightly higher than Case 1 because excessive imports and exports roughly counterbalance.
- Two-month microgrid operation: 84.5% SCR and 747 kWh battery throughput are obtained with the LSTM forecast, with 5141 kWh imports and 111 kWh exports.Despite improved PV and storage utilization, Case 3 has the highest total cost, e40.43, because residual forecast errors cause suboptimal charging and discharging.
- Prediction-horizon effects: 95.2% SCR in Case 1 and 97.9% in Case 3 are reached at H=48, after which SCR declines, particularly for the LSTM scenario.The persistence case remains consistently low and flat across prediction horizons because of limited forecasting capability.
- Prediction-horizon effects: 1200 kWh in Case 1 and 892 kWh in Case 3 are the peak battery throughputs at H=48, followed by slight declines at longer horizons.Throughput increases with horizon length in the ideal and LSTM scenarios, indicating more effective energy shifting from improved PV forecasts.
5. Conclusion
The study concludes that accurate PV forecasting integrated with predictive control improves grid-connected PV-BESS microgrid operation, while control policies must address forecast uncertainty, battery aging, and grid constraints.
- Conclusions: PV forecast accuracy significantly affects microgrid operation, with the LSTM-based approach improving grid-injection reduction, battery utilization, and PV self-consumption over the persistence model.The perfect forecast scenario achieved the best performance across all metrics.
- Conclusions: Higher battery throughput from frequent LSTM-based cycling may accelerate battery aging, requiring control policies that account for degradation and grid constraints.The study highlights battery aging and grid constraints as important considerations for practical control design.
- Limitations: The study assumes perfect load forecasts, fixed battery efficiency, and ideal grid access, limiting how fully its results represent real-world conditions.These assumptions simplify system dynamics and operational conditions.
- Future research: Future research should address joint PV and load forecasting, probabilistic MPC under uncertainty, and ancillary services such as demand response or grid support.These directions aim to improve robustness and extend microgrid functionality.
CRediT authorship contribution statement
The authors contributed across drafting, visualization, software, methodology, conceptualization, validation, supervision, and funding acquisition.
- Berhane Darsene Dimd handled the original draft, visualization, software, methodology, and conceptualization.
- Steve Völler contributed review and editing, validation, and supervision.
- Ole-Morten Midtgård contributed supervision, conceptualization, and funding acquisition.