Segmenting the grid helps prevent cascading failures by 60%

We have proposed a method to reduce the risk of blackouts in electrical power transmission networks by controlling the power flowing through certain lines. These lines can be the existing HVDC lines interconnecting different synchronous areas or new HVDC lines replacing AC lines connecting different zones within the same synchronous area, thus segmenting the power system.

We have used the OPA model to evaluate the changes in the risk of blackouts as a result of the segmentation. The HVDC lines and converters are modeled in this study as lines with controllable impedance. Changing the impedance is a way to control the power flowing through a line in our model. In a cascading failure the control method consist in choosing an appropriate value of the impedance, i.e. the power flow, for each HVDC line to minimize the load shed of the subsequent blackouts. This is done at dispatch by simulating a priory the cascade many times, each with different impedance values, in a Monte Carlo fashion, and selecting the configuration that minimizes the load shed. Our results show that: (i) this method can reduce the risk of blackouts of a homogeneous balanced segmented grid up to 60% with respect to the original full AC network; (ii) the method is particularly efficient in suppressing blackouts affecting more than one zone; (iii) in unbalanced or heterogeneous networks the global effectiveness is reduced, as the risk is determined mainly by the blackouts in the largest zone, however a substantial reduction of the local risk is still obtained in the smaller sub-networks; and (iv) this method is able to decrease the risk of blackouts without increasing its frequency, a counter effect typically found in other control methods.

In a real set up, the simulations to determine the proper value of the power flow should be done in real time during the cascading failure. While this is possible using sufficient computational power, more efficient ways of determining the optimal power flow value should be pursued. Some insight about the values selected by the Monte Carlo method is given by Fig. 7, where a tendency to take larger values of the impedance for larger blackouts can be appreciated, although the variability in the selected values is very large, and in some cases small values for large load sheds are taken. Using machine learning techniques to determine the optimal values of the impedance could possibly improve the velocity of the method.