An Activation Pattern for Tertiary Control Reserves in the Swiss Power System

The successful operation of interconnected power systems requires the balance of electricity generation and consumption. This is performed by frequency control that comprises three level control processes known as primary, secondary, and tertiary control. First, primary control stabilizes frequency within some seconds. Next, secondary control keeps the frequency close to its nominal values and restores scheduled power interchanges to their trade values. Both these control schemes automatically react to power imbalances. Tertiary control assumes a complementary role to secondary control reserves; that is, in the event of persistent outages and saturated secondary control reserves, the system operator manually activates tertiary control reserves based on his experience, taking into consideration that tertiary control power is available with a time delay. There is, however, no common framework for the activation of tertiary control reserves. The frequent activation of tertiary control reserves relieves secondary control reserves, and hence, increases the security of power systems. In addition, manual reserves generally cost less than automatic reserves since they are slower and less flexible. This means a pattern for regular activation of tertiary control reserves would be economically beneficial. This thesis focuses on designing a framework for the activation of tertiary control reserves in order to improve frequency control performance. Considering the fact that tertiary control reserves are not immediately available and, therefore, must be requested in advance, a pattern is proposed based on a Model Predictive Control (MPC) scheme. MPC is a model-based control technique that synthesizes a control decision by performing an optimization procedure. Optimization setup takes operational concerns, such as system security and market conditions, into consideration. Linear regression models based on a robust Maximum Likelihood Estimate (MLE) are employed to predict the control signal. For this purpose, different models based on different distribution functions are tested. The proposed optimization technique is simulated using measurements from the Swiss power system to illustrate technical and economic aspects.