Traffic regulation is an important part of metro networks. Trains in an urban network need guidance to ensure smooth operation of the system, passengers satisfaction, and also to meet criteria fixed by operators or by their clients (usually quality contracts are fixed by local authorities). These criteria can vary from a line to another, but also depending on the day and time of the day. Examples of criteria that have to be met are punctuality, regularity of service, energy consumption... For these systems, forecasts are designed in the form of timetables describing departure and arrival dates of trains at stations, or with promises of regular pace at some stations. However, due to unpredictable delays, these plannings are never met. Such random delays can originate from weather conditions, users misbehavior, signaling systems failure, etc. As a consequence, a metro network cannot be operated without corrective mechanisms called regulation algorithms. Currently, regulation algorithms are mainly event-based and reactive: upon arrival of a train, the difference between the forecast reference arrival date and the actual observed arrival date calls for corrections to the forecast: next departures can be delayed, dwell times can be shortened, commercial speeds can be increased and, in some cases, trains ordering can even be modified. These algorithms are mainly application of rules of the form "if train x is late by more than y seconds, then shorten dwell time by v seconds". Of course, algorithms can be more involved than in this example, and may consider more parameters than a single train delay; yet, they remain quite local decisions. In this setting, one can notice that regulation advices are just application of rather logical rules aiming at recovering delays, but whose optimality is not certain. Experience shows nevertheless that these systems work well in practice and are sufficient to recover from small delays. Nowadays, as the number of daily commuters increases, metro systems have to face growing traffic, while maintaining a high quality of service. In addition to service increase, energy consumption optimization is also becoming a central concern in most cities. There is hence a clear need for algorithms with optimal performance, and for tools to demonstrate this optimality. We propose a framework to model networks that integrate optimization schemes in their regulation , and tools to evaluate the performance of these optimized regulation algorithms. The main idea here is to consider more global decisions: instead of reacting to a local delay, we advocate the fact that solutions returned by regulation algorithms have to consider elements from the whole network and its planned forecast. We also advocate the fact that decisions that are proposed by regulation algorithms have to be optimized: instead of computing a rescheduling that simply postpones forecasts depending on a measured delay and on fixed thresholds, regulation can: • reconsider ordering of trains (and even insert or remove trains from a network)