[Paper Review] Model-free Load Control for High Penetration of Solar Photovoltaic Generation

Ouassim Bara, Mohammad Olama+4 more|arXiv (Cornell University)|Jul 8, 2017

Smart Grid Energy Management34 references11 citations

TL;DR

This paper proposes a model-free control (MFC) framework that enables building HVAC systems to locally consume solar photovoltaic (PV) generation in real time, minimizing grid impact without requiring precise building models. By using data-driven, intelligent proportional controllers, the method tracks PV generation profiles while maintaining indoor temperatures within a ±1°C comfort band, reducing voltage fluctuations and two-way power flow with a tracking error below 1 kW.

ABSTRACT

This paper presents a new model-free control (MFC) mechanism that enables the local distribution level circuit consumption of the photovoltaic (PV) generation by local building loads, in particular, distributed heating, ventilation and air conditioning (HVAC) units. The local consumption of PV generation will help minimize the impact of PV generation on the distribution grid, reduce the required battery storage capacity for PV penetration, and increase solar PV generation penetration levels. The proposed MFC approach with its corresponding intelligent controllers does not require any precise model for buildings, where a reliable modeling is a demanding task. Even when assuming the availability of a good model, the various building architectures would compromise the performance objectives of any model-based control strategy. The objective is to consume most of the PV generation locally while maintaining occupants comfort and physical constraints of HVAC units. That is, by enabling proper scheduling of responsive loads temporally and spatially to minimize the difference between demand and PV production, it would be possible to reduce voltage variations and two-way power flow. Computer simulations show promising results where a significant proportion of the PV generation can be consumed by building HVAC units with the help of intelligent control.

Motivation & Objective

• To minimize the impact of high-penetration solar PV on distribution grids by enabling local consumption of PV generation.
• To maintain occupant thermal comfort within a ±1°C band despite variable solar generation and dynamic building conditions.
• To eliminate the need for complex, precise building models or computationally intensive optimization in real-time control.
• To reduce voltage variations and reverse power flow by synchronizing HVAC load scheduling with real-time PV output.
• To demonstrate a scalable, robust, and low-cost control solution suitable for widespread deployment in residential and commercial buildings.

Proposed method

• The method employs a model-free control (MFC) framework based on the ultra-local model, where only input and output data are used, and physical laws or differential equations are ignored.
• Intelligent proportional controllers are designed to adjust HVAC unit power consumption in real time based on the difference between actual PV generation and current load demand.
• The control strategy enforces a constraint that the total HVAC energy consumption must follow the PV generation profile within a tolerance margin (ε = 1 kW in simulations).
• The approach avoids optimization by directly shaping load demand to match PV supply, using only measurable signals (PV output and indoor temperature).
• The control law is derived from a data-driven formulation that ensures stability and robustness against disturbances such as weather changes and occupancy variations.
• The system is scalable: adding more buildings improves tracking performance without increasing computational complexity or altering the control architecture.

Experimental results

Research questions

• RQ1 Can HVAC systems be controlled in a model-free manner to consume solar PV generation locally, minimizing grid impact?
• RQ2 How can thermal comfort be maintained while synchronizing HVAC load with variable PV output in real time?
• RQ3 To what extent can model-free control reduce voltage fluctuations and reverse power flow in distribution networks with high PV penetration?
• RQ4 Can the proposed method achieve accurate PV profile tracking without relying on complex optimization or precise building models?
• RQ5 What is the effect of increasing the number of buildings on the tracking accuracy and thermal comfort performance?

Key findings

• The model-free control approach successfully tracks the PV generation profile with a maximum tracking error of less than 1 kW, significantly reducing mismatch between supply and demand.
• Indoor temperatures for all 13 buildings remained within the ±1°C comfort band, with only minor violations in the tightest scenarios due to low initial temperatures.
• Adding one more building (N_b = 14) improved temperature regulation and reduced the load on individual HVAC units, demonstrating scalability and performance gain.
• The method maintains robust performance under disturbances such as weather changes and occupancy shifts, without requiring detailed building models.
• The control framework is computationally efficient and scalable, with no increase in computational load when expanding from 13 to 14 buildings.
• The approach effectively reduces two-way power flow and voltage variations by enabling local PV consumption through responsive HVAC loads.