**4.1 Case study**

The area identified for the case study is the city centre of Naples (regional capital of Campania, southern Italy). This area is characterised by a population of 978,399 and a population density of 8220 per km<sup>2</sup> . Moreover, the metropolitan area has a number of inhabitants is around 3,118,000 and the population density is 2645 per km2 . Also, the total number of internal systematic yearly trips is around 685,000.

Two main roads are connected, Via Francesco Caracciolo and Via Riviera di Chiaia, from the West to the East side of the city. Current traffic rules and the connection between these two sides with two concurrent paths are implemented. Two paths are identified, path 1 goes through the Galleria Vittoria, and path 2 is composed of Via Chiatamone, Via Nazario Sauro and Via Acton. The sub-network layout is reported in **Figure 3**.

The network comprises four signalised junctions, among them traffic signals in Section 1, 2 and 5 are pedestrian traffic signals.

In terms of implementations remarks it must be highlighted that the whole traffic control procedure operates every control interval (every five minutes).

To optimise the traffic signal decision variables, the rolling horizon approach is adopted combined with a traffic flow prediction model. In particular, the rolling horizon itself is characterised by two terms the roll period (equal to five seconds) and the look ahead period (starting at the end of the roll period and ending at the upper bound of the prediction); in order to further guarantee the consistency with the traffic flow, traffic information are collected every roll period.

It must be distinguished that traffic signals in Section 1 and 2 are managed through LM whereas traffic signals in sections 4 and 5 are optimised through NTC. In general, the duration of the cycle length is 110 seconds and the stages 1 2 and 3 are respectively equal to 19 seconds, 65 seconds, and 26 seconds.

Regarding the origin - destination flows (and then the entry exit matrix definition) these have been obtained by combing the results of a macroscopic static traffic assignment procedure (PUMS - NAPOLI) [20] with a traffic counts survey done in 2017; in

**Figure 3.** *Topology of the tested network.*

particular traffic counts were collected at the beginning of 2017 in two peak hours of the day (mooring, from 7. 30 until 10.30 and afternoon, from 17.30 until 20.00).

The traffic flow was microscopically model through SUMO which is able to guarantee the on-line consistency of the procedures by adopting the TraCI interface, and the supporting code was developed in MATLAB (the R2018b was adopted). The input of the optimisation procedures are the travel times (TT) and the queue lengths (QL) that are collected through specific detectors located on the network. Due to the stochastic nature of the microsimulation approach, each simulation is run twenty successive times and the final values are provided by averaging the value of each simulation.

In terms of the goodness-of-fit function for model calibration, the Geoffrey E. Havers statistic [21] was adopted considering the observed and modelled data and the correspondence is less than 5 for 75% of the pairs (in accordance with the guidelines provided in [22].
