Intellectual Merit: A high penetration of renewable energy sources in the power system is a major component of the vision for a clean and sustainable electric power generation. However, the main target in terms of renewable generation is currently on wind generation. The intermittency of these sources as well as the fact that the locations of high availability of the primary energy source wind do not coincide with the locations of the load center are major challenges which have to be resolved. For a successful integration of wind generation, we must find a way to balance this intermittency effectively and to transmit the power to the loads without the need for environmentally unfriendly backup generation and a substantial extension of the transmission system. In this project, these challenges are addressed by proposing a new distributed predictive control algorithm for the coordination of the intermittent energy sources with storage, demand control and backup generation. The proposed predictive control allows for an optimal utilization of the available capacities with the objective to minimize the overall use of backup generation but also the changes in its output as well as minimizing the required demand control. The control is carried out in a distributed way taking into account that control of the power system is shared among several entities. In addition, the question is asked how much storage and backup generation capacity and load flexibility is needed for a reliable operation of the power system with a given level of intermittent renewable power generation penetration and vice versa. This is a fundamental question for the planning of the future power system. A systematic tool based on stochastic programming is proposed to answer this question and to give an estimate for the feasibility of the intended intermittent generation penetration.
Broader Impacts: The resulting problem size and the involved computation effort or the shared control structure so far often prevented an application of predictive control to large-scale systems. The features of the proposed distributed predictive control are low computational effort, fast convergence and no parameter tuning which make this control algorithm applicable to large-scale systems and superior to other distributed predictive control methods. Even though the distributed predictive control method is proposed for power systems, the mathematical framework is applicable to many other large-scale problems opening the door for predictive control to other large scale systems. The results of this project will be included in the content of existing courses providing access to research results for students.
A high penetration of renewable energy sources in the power system is a major component of the vision for a clean and sustainable electric energy future. However, the main target in terms of renewable generation is currently on wind generation. The intermittency of these sources as well as the fact that the locations of high availability of the primary energy source wind do not coincide with the locations of the load center are major challenges which have to be resolved. For a successful integration of wind generation, we must find ways to balance this intermittency effectively and to transmit the power to the loads without the need for environmentally unfriendly backup generation and a substantial extension of the transmission system. Intellectual Merit: In this project, contributions to the solution of these challenges were made by the development of two main tools: (1) a method to coordinate resources located in multiple control areas without the need for the areas to exchange significant amounts of information and (2) an approach for optimal sizing of storage devices in an environment in which the power output from some of the generation resources are uncertain. For the coordination method, particularly the situation in which one area has significant amounts of renewable generation and the other has significant amounts of storage was considered. The coordination of the areas was accomplished by using the Optimality Condition Decomposition in which each subproblem corresponds to one of the control areas. Two extensions to that method have been made which are both based on exchanging a little bit more information compared to the original problem resulting in faster convergence of the distributed algorithm to the overall solution. The optimization problem that was decomposed is a model predictive control problem, i.e. a multi-step optimization problem which ensures optimal usage of the storage. For the storage sizing problem, a tool was developed which is based on stochastic optimization and takes into account a wide range of possible evolutions of load and renewable generation over the day. A scenario reduction technique was derived based on the discovered fact that the optimal storage size is directly related to the variance in the marginal generation cost in the system. Hence, this allows reducing the very large problem size to a more manageable level. Furthermore, probabilistic constraints were incorporated into this problem to reflect that there are forecast errors in renewable generation and that these need to be compensated. Hence, the forecast error is modeled with a probability distribution and the probability constraints ensure that the system is able to balance the error with a certain pre-specified probability without having to shed load or curtail renewable generation. Moreover, the proposed formulation also accounts for the fact that the system may be operated under a receding horizon concept which is important as it allows for feedback and should result in a decrease in the required storage size. Broader Impact: The features of the developed distributed predictive control are low computational effort for each individual subproblem, fast convergence and no parameter tuning which make this control algorithm applicable to large-scale systems. Even though the distributed predictive control method has been applied to power systems in this project, the mathematical framework is applicable to many other large-scale problems opening the door for predictive control to other large scale systems with shared control responsibility. With the proposed storage sizing tool, it is possible to determine what storage size is optimal, i.e. cost-effective, for a given prediction error distribution function and existing generation portfolio. This will have both short-term and long-term impacts on power systems planning and control and will help making optimal planning decisions on the path towards a sustainable energy future. The results of this project have been included in the content of courses providing access to research results for students. It has led to a general increase of interest in optimization and power systems in the students that attended the class. In addition, graduate students have done the bulk part of this work which gave them experience and knowledge which is currently highly sought due to an industry with an aging workforce.
