Observability and Decision Support for Supervision of Distributed Power System Control

Analyses of past disturbances and blackouts have shown that lack of situation awareness of control room operators led to wrong or suboptimal decisions. Two main issues are associated with lack of situation awareness. On the one hand, operators do not get the information needed to fully understand the current state of the grid. On the other hand, the information is not correctly interpreted. Along with the enhancement of existing tools, complementary monitoring methods must be developed for increased observability and situation awareness. To facilitate the decision process, tools that provide specific operational information, relevant to the current grid condition, need to be developed. This dissertation covers three areas where specific challenges for improved observability and decision support in future control rooms are addressed: Classical large power system stability issues, innovative data-driven techniques for voltage estimation at the distribution level, and
aggregation functions for geographically dispersed time-dependent distributed energy resources are investigated. In particular, this dissertation proposes a decision support tool for transient stability preventive control, a neural-network-based approach for real-time voltage estimation in active distribution grids, and a modeling approach to harness the flexibility of an aggregation of electric vehicles. For improved monitoring and maintaining power system stability, a decision support tool for transient stability preventive control, based on time-domain simulations, is proposed. The approach employs a critical bus screening and a fast critical contingency screening prior to the assessment to reduce the computational burden. In addition, a fast-converging technique to determine the required dispatch for re-establishing a predefined stability margin is presented. The approach delivers a near optimal solution in terms of cost minimization due to its sequential nature and shows to be robust when applied to larger power systems. The general problem of low distribution grid observability is addressed by proposing a data-driven
approach for real-time voltage estimation in active distribution grids utilizing existing real-time and non-real-time measurements. The framework to establish such a neural real-time voltage estimator is described. Then, the capabilities and limitations of the approach under practical considerations are analyzed by means of field data. This includes a methodology to select the most relevant input variables and find a tradeoff solution between achievable performance and number
of inputs and a sensitivity analysis of the performance and number of neurons. Moreover, the quantity of historical data which is needed to train an adequately functioning model is analyzed. In order to maintain the accuracy of the trained model, the impact of the retraining interval on the model performance is determined. Additionally, the sensitivity of the model to distributed generation is investigated. The growth of electric vehicles results in additional demand for charging which will require large investments in power distribution, transmission, and generation. However, this demand is often also flexible in time and can be actively managed to reduce the required investments and to enhance power system operation. Harnessing this flexibility requires forecasting and controlling electric vehicle charging at thousands of stations. A model to aggregate and forecast the flexible demand from tens to thousands of electric vehicle supply equipments is proposed. From an
operational perspective, the aggregated EVs are represented as an equivalent time-variant storage model whose parameters can be easily aggregated and forecasted using autoregressive models. The forecastability of the uncontrolled demand and the storage parameters is evaluated using an extensive dataset from 1341 non-residential electric vehicle supply equipments located in NorthernCalifornia. Two possible applications of the model are presented: peak reduction compared to
uncontrolled charging, and an energy arbitrage scenario. Overall, it is shown that a combination of classical and innovative approaches can contribute to improved situation awareness of control room operators. In addition, complementary decision support tools can further aid the control room operator in determining appropriate operational actions. However, observability and decision support tools must be integrated to allow optimal operational decisions that satisfy all constraints.