Introduction
In modern smart homes equipped with photovoltaic (PV) systems, a significant challenge arises from the "temporal mismatch" between energy production and consumption. Solar generation typically peaks at midday when residents are away, while energy demand surges in the evening. While chemical batteries can bridge this gap, their high cost remains a barrier. This research explores an alternative: using standard household water heaters as "thermal batteries" to store excess solar energy as heat.
System Architecture and Problem Formulation
The core of the proposed solution is an Energy Management System (EMS) that coordinates energy flows between the PV generator, the electrical grid, and household demands. Unlike electrical batteries, thermal storage in water heaters is non-reversible; once electricity is converted to heat, it cannot be recovered as electricity. Furthermore, optimal control is complicated by the inherent uncertainty in weather forecasts, electricity prices, and resident behavior.
Methodology: Stochastic Dynamic Programming
To address these uncertainties, the study employs Stochastic Dynamic Programming (DP). Unlike deterministic models that assume perfect foresight, the stochastic approach treats prediction errors in weather, demand, and pricing as random variables (modeled with Gaussian distributions).
The DP algorithm works backward from a future state to determine the optimal heating schedule that minimizes the total "cost-to-go." By calculating the probability of reaching specific temperature states under uncertain consumption, the system can make more robust decisions. This method is compared against traditional PI (Proportional-Integral) controllers and deterministic DP models.
Key Insights: The "Buffer" Effect
A critical finding is the emergence of a "buffer" effect in the stochastic model. Because the algorithm accounts for the risk of unexpected hot water usage, it naturally "pre-heats" the tank when solar energy is abundant or grid prices are low. This ensures that the water temperature remains above the comfort threshold (60°C) even if forecasts prove inaccurate.
In contrast, deterministic models often "gamble" too aggressively on perfect forecasts. When a resident uses more water than predicted, these models fail to maintain the required temperature, leading to comfort violations.
Results and Performance
The performance of the Stochastic DP was evaluated over a simulated three-day window using real-world data from Munich, Germany, and ComEd pricing.
- Reliability: The Stochastic DP achieved zero comfort violations. In comparison, the Shortest Path and Deterministic DP models suffered from 179 and 123 violations, respectively, when no corrective heating was applied.
- Cost Efficiency: The Stochastic DP reduced total costs to 8.77€, compared to 9.08€ for the baseline PI-controller. While deterministic models initially appeared cheaper, their costs rose significantly once emergency heating was factored in to correct temperature drops.
Conclusion
By leveraging Stochastic Dynamic Programming, household water heaters can be effectively transformed into reliable thermal storage units. This approach maximizes the use of self-generated solar energy and minimizes electricity costs without sacrificing resident comfort. Future research will focus on expanding the model to include more complex tank physics (stratification) and integrating other household loads like HVAC systems and electric vehicle charging.
Summary updated