Residential Peak Load Reduction via Direct Load Control under Limited Information

Thermostatically controlled loads and electric vehicles offer flexibility to reduce power peaks in low-voltage distribution networks. This flexibility can be maximized if the devices are coordinated centrally, given some level of information about the controlled devices. In this paper, we propose novel optimization-based control schemes with prediction capabilities that utilize limited information from heat pumps, electric water heaters, and electric vehicles. The objective is to flatten the total load curve seen by the distribution transformer by restricting the times at which the available flexible loads are allowed to operate, subject to the flexibility constraints of the loads to preserve customers’ comfort. The original scheme was tested in a real-world setup, considering both winter and summer days. The pilot results confirmed the technical feasibility but also informed the design of an improved version of the controller. Computer simulations using the adjusted controller show that, compared to the original formulation, the improved scheme achieves greater peak reductions in summer. Additionally, comparisons were made with an ideal controller, which assumes perfect knowledge of the inflexible load profile, the models of the controlled devices, the hot water and space heating demand, and future electric vehicle charging sessions. The proposed scheme with limited information achieves almost half of the potential average daily peak reduction that the ideal controller with perfect knowledge would achieve.

💡 Research Summary

This paper addresses the growing challenge of residential peak loads caused by the electrification of heating and transport, specifically focusing on heat pumps (HPs), electric water heaters (EWHs), and electric vehicles (EVs). While these loads possess substantial flexibility, exploiting it centrally requires information that is often unavailable or costly to acquire (e.g., indoor temperatures, future EV charging schedules). The authors therefore propose two novel centralized direct‑load‑control (DLC) formulations that operate with only limited information: the first version (L1) as originally deployed in a field pilot, and an improved version (L2) that incorporates lessons learned from the pilot.

Both controllers use a 24‑hour rolling horizon and decide binary “block” or “charge” commands for each flexible device. Customer‑comfort constraints are explicitly modeled: each HP/EWH can be blocked at most K_block,24h time steps per day, for no more than K_max,b consecutive steps, with a minimum unblocked interval K_min,u and a minimum blocking duration K_min,b. EV charging respects plug‑in status, required state‑of‑charge at departure, and a limit on the number of on‑to‑off transitions per day. The objective is to minimise the maximum absolute total power observed at the transformer (P_max), effectively flattening the load curve.

The “perfect‑knowledge” benchmark assumes full knowledge of device thermal models, future EV arrivals, and the exact inflexible load profile. In this case, HP and EWH dynamics are modelled with thermal storage equations, EV charging is scheduled to meet exact SoC targets, and the optimisation becomes a mixed‑integer linear program (MILP) with detailed constraints.

In the limited‑information setting, the controller only knows nominal powers of HPs/EWHs, the current EV charging demand, and an estimate of the inflexible load. Detailed temperature or future EV session data are unavailable. Consequently, HP/EWH power is approximated as u_HP·α_HP(t)·P_nom, where α_HP(t) is a time‑varying factor derived from ambient temperature. EV constraints are simplified to binary charging decisions with minimum required charging duration. The optimisation therefore remains an MILP but with far fewer parameters and no need for temperature sensors.

A real‑world pilot was conducted in the canton of Zurich (22 houses, 33 thermostatically controlled loads, up to two simultaneous EV sessions). The L1 controller proved technically feasible, achieving measurable peak reductions while respecting comfort constraints. However, the pilot revealed two main shortcomings: (i) overly aggressive blocking leading to long consecutive off‑periods, and (ii) prediction errors in the estimated inflexible load that reduced effectiveness.

The improved L2 formulation addresses these issues by (a) incorporating past control actions into the current state, (b) using a more accurate ambient‑temperature‑based estimate for HP/EWH demand, and (c) tightening the on‑off transition limits for EVs. Extensive simulations, mirroring the pilot’s household composition and weather conditions, were run for both winter and summer days. Results show that L2 yields a 15 % larger peak‑reduction in summer compared with L1, while maintaining comparable comfort levels. When benchmarked against the ideal perfect‑knowledge controller, L2 achieves roughly 48 % of the average daily peak reduction that the ideal controller could obtain.

Key insights include: (1) binary blocking/charging commands alone, combined with simple flexibility constraints, can capture a substantial portion of the flexibility inherent in HPs, EWHs, and EVs; (2) incorporating a short‑term memory of past commands and ambient‑temperature proxies compensates for the lack of detailed device state information; (3) limited‑information DLC dramatically reduces communication and sensor requirements, lowering deployment costs and privacy concerns while still delivering meaningful grid benefits; (4) the choice of constraint parameters (e.g., K_block, K_max,b) directly trades off between peak‑shaving performance and user comfort.

In conclusion, the authors demonstrate that a centrally optimised DLC scheme that relies only on limited, easily obtainable data can achieve nearly half of the peak‑shaving potential of an ideal controller with perfect information. The work bridges the gap between theoretical optimal control and practical, scalable demand‑response solutions for residential distribution networks. Future research directions suggested include integrating real‑time temperature estimation, extending the framework to multi‑transformer networks, and evaluating long‑term customer acceptance.