On the three-step control methodology for Smart Grids

Energy usage, dependability of energy supply and climate change are important topics in society. Today, a number of trends can be recognized in the electricity consumption and generation. On the one hand, the electricity consumption increases and becomes more fluctuating. This is caused by increasing economic activity and prosperity, but also by a shift towards more electricity supplied devices, for example electrical cars. On the other hand, the projected reduction in CO2 emission requires the introduction of generation based on renewable sources. These renewable sources are based on uncontrollable and very fluctuating sun-, water- and wind power. These trends introduce challenges to maintain a reliable and affordable electricity supply. Fluctuations in electricity demand decrease the efficiency of conventional power plants. Furthermore, peak demands determine the required generation and transportation capacity, higher peaks require more generation and transportation capacity. Moreover, renewable generation often does not match the demand profile and in addition to that inefficient peak power plants have to supply this fluctuating mismatch between generation and demand. Furthermore, since all produced electricity must be consumed, in the current electricity supply system the lowest demand results in an upper bound on the renewable generation. Therefore, only a limited amount of the demand can be supplied by renewable sources. A solution to these problems may be to transform domestic customers from static consumers into active participants in the production process. Consumer participation can be achieved through the development of new (domestic) devices with controllable load, micro-generation and domestic energy storage of both heat and electricity. These devices have potential to shift electricity consumption in time without harming the comfort of the residents. Examples of devices with optimization potential are (smart) freezers and fridges which can adjust their cooling cycles to shift their electricity load and (electrical car) batteries that can temporarily store excess electricity. To exploit this optimization potential on a large scale, a global control methodology is required. In this thesis the above mentioned challenges and optimization potential are studied and a control methodology is derived. This con- trol methodology aims 1) to achieve a more efficient use of the generated electricity of existing power plants, 2) to facilitate the large scale introduction of renewable sources and 3) to allow large scale introduction of new technologies for production, consumption and storage while at the same time maintaining grid stability and ensuring a reliable and affordable supply. In this thesis, first a model of the energy infrastructure is derived. This model consists of multiple levels: the leaves are the devices within buildings, modelling a building as a collection of devices each with their own behavior and optimization potential. These devices can convert, buffer and consume energy and are connected to each other in such a way that energy can flow between these devices. Buildings can exchange energy with their outer world and multiple buildings can be combined into a grid. The electricity grid also consists of multiple levels, from the low voltage distribution level up to the high voltage lines connected to power plants. The different levels are connected by transformers, which are converting devices with their own characteristics (e.g. capacity). The low voltage distribution level can be split up in multiple segments, each with a number of buildings connected. These individual segments model a neighborhood, whereas multiple neighborhoods can be combined into a city, etc. The core of the developed control methodology is a three-step control methodology introduced in this thesis. The goal of this control methodology is to tackle the above mentioned challenges by exploiting the domestic optimization potential of domestic customers. Domestic is in this case broader than only houses, it comprises all consumers at the distribution level: houses, schools, shops, factories, etc. The control methodology is based on a three-step approach to monitor and manage domestic energy demand as well as the generation and storage of the energy. In the first step, the energy usage or production for every individual building in the grid is predicted on a device level. In the second step, the predictions of the individual buildings are aggregated and a global planning is made in an iterative and hierarchical way. The result of this planning is an energy profile. In the last step, a local scheduler in every building schedules the devices in realtime, using steering signals determined by the global planning as an input. The base of all three steps of the control methodology is the mathematical analysis of the energy streams expressed in the energy infrastructure model in combination with mathematical optimization techniques. The main focus of this thesis is on the three-step control methodology and in particular on the last step of this control methodology. The aim of the third step of the proposed control methodology is to decide which devices are to be switched on and when. The main task of the realtime controller of the third step is to guarantee the comfort level of the residents. Within the boundaries of this comfort level, it can exploit scheduling freedom to work towards certain objectives. The control methodology can use the steering signals from the global planning as input, but may optionally also incorperate realtime inputs. The overall control methodology can have two types of optimization objectives: 1) it can work towards a predefined consumption profile (e.g. lower peaks) or 2) it can react on realtime fluctuations (e.g. caused by renewable sources). The planning determined by the second step of the control methodology is based on the predictions of the energy usage and behavior of customers, expressed in predictions of when and for how long devices are running (runtime). Based on these predictions and the objective of the optimization, a planning for the runtimes of devices can be derived. However, the actual behavior often deviates from the predictions (prediction errors). Due to these prediction errors, the actual situation often deviates from the predicted situation and it is not possible to stick to the planning. Furthermore, the steering signals do not always fit with the current situation anymore and can result in decisions that are very disadvantageous for later periods. The last step has to try to work around these prediction errors while keeping the closest possible to the planning. Therefore, the realtime controller is extended with Model Predictive Control (MPC). Within a Model Predictive Control approach the realtime controller not only takes the current situation into account, but also a certain period in the future. Short-term predictions of the behavior of the devices are made and based on the current situation, the short-term predictions and the steering signals, the best decision for the current situation is taken. The decisions of the realtime controller are based on cost functions. The preferences and optimization potential for every device are expressed using cost functions. These cost functions express the desirability of possible choices. Furthermore, the steering signals of the global planner are also expressed as cost functions. Based on the cost functions and (technical and social) constraints the best decisions can be determined by the realtime controller. Cost functions are a very generic way of expressing preferences and optimization potential of devices and technologies. However, it is important to define the cost functions correctly, otherwise undesirable and unpredicted behavior can occur. Based on the model of the energy infrastructure a simulator has been developed to simulate future energy infrastructure scenarios and control methodologies to tackle earlier mentioned challenges. The simulator is based on discrete simulations and the design is kept the closest possible to the energy infrastructure model. Individual buildings are specified on a device level and multiple buildings are combined into a (Smart) Grid. The simulator is fast, it can simulate a large group of buildings within reasonable time. Furthermore, it is generic in the sense that it allows to simulate a lot of different scenarios, technologies and control methodologies. The modular structure of the simulator eases the addition of models of new technologies and due to initiation through configuration files it is easy to define (large) scenarios. The extensive logging capabilities of all elements of the simulator enables the possibility to study the simulation results in detail. The possibility to use real world data for the simulations and the ability to use a broad range of stochastic variations helps to define a realistic scenario with only a limited amount of input data. To study the effectiveness of the control methodology, to find the best parameters of the control methodology and to study the most economic use of the flexibility of devices, multiple scenarios have been simulated. The simulations show that the control methodology can optimize the energy flows and can control the operation of the domestic devices in an economic manner without discomfort for the residents. The prediction and planning step preceding the realtime control improves the results of the control methodology: much more optimization potential can be exploited and the results are more predictable and dependable. Furthermore, the addition of MPC strengthens the capabilities of the realtime controller to work around prediction errors. Prototype experiments show that it is possible to incorporate the optimization algorithms in a real prototype and that the algorithms are able to manage the behavior of the devices within the comfort levels. Furthermore, experiments show that a microCHP device can act as a backup generator in case of a power cut: a disconnected grid situation was created using a microCHP device, a battery and multiple devices controlled by the optimization algorithm. Based on the simulations and prototype experiments we conclude that the control methodology can monitor and adjust the consumption profiles and electricity streams. It can use the optimization potential of a large group of buildings to work towards global objectives. However, it requires a change of mind of customers to give the control of their devices to a global controller. The combination of prediction, planning and realtime control is a promising direction for control methodologies for Smart Grids: it is scalable, generic and reliable. The three-step approach can provide an important contribution to realize the European 20-20-20 ambition [22].