Actieve vraagsturing van clusters huishoudtoestellen via gereduceerde-ordemodellen: brug tussen theorie en praktijk

Large-scale deployment of renewable energy sources is seen as a main contributor towards the transition to a more secure, affordable and decarbonised energy system. These intermittent and unpredictable energy sources however lead to variable and difficult to predict bidirectional power flows in distribution and transmission grids. Simultaneously, the total electricity consumption is increasing as opportunities for electrification of transport and heating arise. These trends are leading to increasingly complex supply and demand power flows, pressurising the electricity grids, markets and actors that are currently in use. In recent years demand response has received renewed research attention due to advances in control theory and the abundance of affordable computational and communication resources. Harnessing the flexibility on the residential demand side to help maintain the delicate balance between production and consumption does however pose several challenges. First, the vast number of communicating devices requires a cost-competitive solution. Secondly, the development of an efficient, practical, and reliable control scheme poses a significant challenge due to the large dimensionality of the problem. Next, the comfort of the end-user should always prevail. Also, the future smart electricity grid will consist of many stakeholders, each with their specific and opposing objectives. The final challenge consists of evaluating the most suitable control methodologies, demonstrating their usefulness in simulated and physical environments. This dissertation is focussed on the development of efficient, practical, and reliable control approaches for large clusters of residential appliances. The resulting virtual power plant can be used to offer valuable services to utilities and system operators, transforming end-users into green, active and profitable stakeholders of the energy transition. To this end, an existing market-based control approach has been predominantly adopted, extended and modified. Its inherent drawbacks of imperfect state information due to aggregation, dispatch deviations resulting from the bid functions, and applicability to thermal electrical appliances have been mitigated or resolved. Throughout this work, the economic and technical feasibility of the approaches was strongly kept in mind, and supported by practical, real-world deployments. The first part of this work uses a heterogeneous mix of device classes as used in the LINEAR project, a large-scale research and demonstration project using flexible appliances of 240 Belgian families. In order to assess the effectiveness of the developed approaches, a portfolio of demand response algorithms has been implemented in a simulation framework. The integrated load-flow solver allows to evaluate the economic and technical impact of realistic scenarios. First, a novel distributed voltage control mechanism is presented that has also been deployed in the pilot. Next, this mechanism is combined with an adaptation of the market-based approach towards appliances used in the pilot. Simulation results show that from an economic welfare perspective, multiple stakeholder objectives should be simultaneously taken into account. Following this, the double-layered control approach is extended with a self-adaptive machine learning technique. By incorporating the non-linear cluster dynamics and local grid congestions, the energy cost can be further reduced. Finally, in order to bridge the gap between highly realistic simulations and a hard-wired roll-out, a deployment architecture has been developed. This allows to rapidly evaluate framework code on a physical laboratory environment, real-time digital simulator, and potentially also pilot projects. The second part is dedicated to thermostatically controlled loads, which offer the largest combined residential power and energy storage capacity. First, the market-based approach is modified towards control of heterogeneous clusters of these appliances. The novel aggregate model based on virtual tracer devices is identified in a non-intrusive manner, and captures both steady-state and transient population dynamics, as well as the heterogeneity of the cluster. Additionally, the dispatch mechanism is included in the optimization, further improving the tracking performance. Finally, a hard-wired field trial has been conducted using 25 student refrigerators in dorm rooms on the UC Berkeley campus. The developed local control modules, communication platform and control methodology simultaneously address the aforementioned barriers.1 Introduction 1.1 Smart Grids 1.2 Demand Response 1.3 Flexible Appliances 1.4 Challenges of Residential Demand Response 1.5 Research Question and Objectives 1.6 Scope of the Work, Outline and Main Contributions 2 Demand Response Simulator 2.1 Agent Based Structure 2.2 Demand Response Algorithms 2.3 Demand Response Programs 2.4 Data Models 2.4.1 Wet Appliances 2.4.2 Electric Vehicles 2.4.3 Electric Water Heaters 2.4.4 Household Load and PV Profiles 2.5 Load Flow Calculations 2.5.1 Grid Topologies 2.6 Communication Effects 2.7 Conclusion 3 Local Voltage Control using Mixed Device Classes 3.1 Related Work 3.2 Pilot Architecture 3.3 Droop Mechanism 3.4 Simulation Scenario 3.5 Simulation Results 3.6 Conclusions 4 Cluster Control of Mixed Device Classes 4.1 Three Step Approach 4.1.1 Step 1: Aggregation 4.1.2 Step 2: Optimization 4.1.3 Step 3: Real-time Control 4.2 Double Layered Control 4.3 Simulation Scenario 4.4 Simulation Results 4.5 Conclusions 5 Data-driven Control: Accounting for Modelling Inaccuracies and Grid Congestions 5.1 Markov Decision Processes 5.2 Problem Formulation 5.3 Model-based Reinforcement Learning 5.4 Evaluation 5.5 Conclusions 6 Rapid Algorithm Deployment Platform 6.1 Introduction 6.2 Algorithm Deployment Options 6.3 Architecture Overview 6.4 Hardware-in-the-loop Scenario 6.5 Simulation Results and Measurements 6.6 Conclusions 7 Cluster Control of Thermostatically Controlled Loads 7.1 Introduction 7.2 Related Work 7.3 Three Step Approach 7.3.1 Step 1: Aggregation 7.3.2 Step 2: Optimization 7.3.3 Step 3: Real-time Control 7.4 Three Step Approach for TCLs 7.4.1 Step 1: Aggregation 7.4.2 Step 2: Optimization 7.4.3 Step 3: Real-time Control 7.5 Resultsnrpages: 166status: publishe