Grid segmentation solves

To assist real time operators in our control rooms, in this emerging context and maintain a robust power supply with high electrical energy quality, new solutions must be devised. They must adapt to decentralized intermittent production and allow for sufficient fast response to manage rapid flow changes.

Autonomous controllers are widely used in the grid, especially for power balance using frequency regulation or voltage/reactive power control [4]. Their use will progressively extend to other fields since they have a crucial role in meeting the challenge of efficiency, ensuring reliable operation, lowering costs, and reducing environmental impacts [4], [5]. In this study, we are interested in power flow control. Operators frequently use topological actions as remedial action for power flow control because they are less expensive than redispatching. Hence we are interested in topological autonomous controller. Topological closed-loop control that consists of controlling the grid topology (powerline and busbar breakers’ statutes) based on grid state monitoring can efficiently maintain a reliable power supply. For highly meshed grid, the number of actions that the agent can perform at each time step is dramatically high for the entire grid. We must therefore think about a more suitable design.

As mentioned by [6], segmenting the power system into finer blocks by creating smaller areas whose operations can be considered independent from each other in various aspects allows a more straightforward assessment of local disturbances and an optimal parallelization. We show in this paper how this parallelization can make network operations more robust and efficient.

Therefore, a decentralized control approach could use multiple local closed-loop controllers (LCLCs) to manage overloads created by rapid, unexpected variations of intermittent renewable productions. We study the interest of using topological remedial actions without considering the possible curtailment of intermittent renewable productions. Thus, the decentralized control technique consists of topological local closed-loop controllers (T-LCLCs). The T-LCLC manages local power transmission distribution and prevents local overload by performing local topological actions from local information of the grid state, allowing fast closed-loop actions to manage sudden power flow variations.

Historically, large grids have been divided into zones to help planning and control applications. There are seven regional dispatching centers in France, each of which manages restricted parts of the grid. The implementation of LCLC on the network is an extension of this idea.

Thus, the solution considered is to divide the network into sub-regions characterized by strong internal influences and weak inter-zone influences in terms of topological actions. These areas must be characterized by measuring the impacts of the local actions inside a zone on the external zones and, reciprocally, the impact of external actions on this zone. It is also necessary to ensure that decentralized parallel optimization maintains a similar quality as centralized optimization while providing robustness and flexibility benefits. Finally, since the arrival of more autonomous agents on the network that perform more actions will increase their influence, making it more cyber–physical, the method used to divide the network should transcribe the cyber–physical effects.

Many existing and future power system control and protection functions use the concept of zones or regions [7], [8]. Power grid segmentation showed efficiency for different applications, such as generator coherency or coherent network areas [9], [10], [11], wide-area voltage control [12], [13], [14], or controlled islanding [6], [15], relying on different methods. Each partitioning is adapted to a specific application and relies on a specific method. In our case, the purpose is to define zones for T-LCLC. Following the same perspective as [6], [16], our segmentation is based on the partition of a sensitivity matrix that represents impacts of a power line switching on power flows of other power lines. This method is adapted to our context as it can take into account cyber–physical effects. We also show that it is necessary to characterize the clusters obtained, mainly by measuring the inter-cluster interactions, to determine the areas in which an automaton can be implemented.

The article is organized as follow. In Section 2, we will describe the functioning of the simulated automaton. In Section 3 we develop the method pipeline to select zones to host T-LCLC. Section 4 is dedicated to the results of the segmentation for the zonal topological agent controller.