Requisitos para ejercer el Comercio

The Legacy Alternative has been created in order to attest the usefulness of the former one. The whole concept of the Proposed Alternative in Section4.1was aimed at pro-portioning the ability to consider secondary offsets, and to enforce offsets in general in a way that would disturb the least possible the traffic conditions. Given the limi-tations of the traffic flow model used (see Section3.4.1 Traffic Flow Model), it could be possible that it would not be capable of reproducing the disturbances caused by

the changing cycle lengths during the process of implementing new offsets. If this was the case, then the Proposed Alternative could even cause more harm than ben-efit comparing to TUC’s original strategy (one cycle transition plans), and perhaps a much simpler approach would be better.

The first point to tackle was the one cycle transition plans. Instead of redesigning and extending the Green Control Module, the Legacy Alternative only actuates in the Off-set Control Module, keeping the rest of TUC’s configuration, as presented in Chapter 3, unaltered. Following the discussion in Shelby et al.2006and the offset configura-tion in Pohlmann2010, the Offset Control Module was modified to limit the transiconfigura-tion cycles to±20%of the current common cycle length, and to keep generating transition plans until the desired offsets were reached, i.e. implementing the Shortway method.

In order to deal with secondary offsets, the same approach presented previously in Section4.1.3 Unexpected Setbackis used. The secondary offsets are “transformed”

into primary offsets, and depending on the subtype of offset control used (Selfish, Democratic, or Partially Democratic), a final offset is calculated for each pair of inter-sections. The transition cycles are then calculated successively as described in Sec-tion3.3 Offset Control Modulein Chapter3, and passed as fixed values for the Green Control Module.

All ATCSs need some sort of traffic state estimation to be capable of determining the right set of actions that will produce a better operation of the traffic network. For the TUC strategy, it is expected the availability of three types of magnitude: queue size; traffic demand; and turning rate. Even though TUC (Diakaki1999) has been presented with an estimator for the number of vehicles inside the links, which dou-bles as queue estimator, a new alternative has been developed and used throughout the current investigation.

In the process of analysing how TUC’s Green Control Module functions, it was identi-fied a possible way to improve its performance. TUC’s strategy aims the control of the network by trying to respond to the stochastic behaviour of traffic, where the queue lengths and vehicle flows change at each new cycle. In order to justify the frequency at which TUC’s control actions are calculated, these changes must be captured. Since these control actions are to be applied on the next cycle, then the simple estimation of the past queue lengths, or even the current amount of vehicles in each link, becomes a suboptimal solution. The ideal scenario would be to predict the queues and, with these future values, calculate the necessary amount of green for each of them.

Looking at a traffic network, it is possible to visualize that the traffic demand of a given link is made up of the traffic flow being let through the upstream links. In the same way, the queues being formed are also a result of the queues that were served by the upstream links. Therefore, if the dynamics of the network, along with the estimated past queue lengths, are taken into consideration, it is possible to better assess how the future queues will develop. And this was the path followed.

The literature presents many alternatives for the estimation of traffic states in urban roads, but none of them really takes advantage of the use of the inherent dynamics of the whole network in order to improve the estimation. There is a range of math-ematical tools being used: Markov Chains (Viti and Zuylen2004; Yu et al.2003);

Neural Networks (Zheng et al.2006); Kalman Filtering (Gang et al.2007; Jabari and

Liu2013; Mück2002b; Pueboobpaphan and Nakatsuji2006; Tampère and Immers 2007); and Cell Transmission Model (Chen et al.2010; Gang et al.2007; Hu et al.2010;

Jabari and Liu2012; Tampère and Immers2007), which confer good results to the solution, but do not directly fit the requirements of the current application. Except the recent work from Jabari and Liu, they either need a much larger interval between each estimation, or just limit the scope to individual road stretches, which in most cases is due to their complexity. Seeking a much simpler approach, Vigos and Papa-georgiou2010presented a queue estimator much similar to the one in Diakaki1999, where individual queue lengths are a direct product of the occupancy of single loop detectors situated in the middle of the links.

From the above cited works, Tampère and Immers2007is the one that better approx-imates the envisioned solution. It uses the CTM as a model for the whole network, and, by mirroring the information from detectors into the related cells in the model, it is capable of furnishing the desired traffic states. The disadvantage of this tech-nique is that a much refined traffic information is needed, so that its resolution in time matches the one in the model, e.g. detector occupancy with a one second reso-lution. Apart from that, the estimation ends up being restrictive in its effectiveness, because only the state estimates of the cells neighbouring the one associated to the real physical detector really profit from it.

Even though the proposed solution resembles the work from Tampère and Immers, it was actually initially inspired by Mück2002b. Mück divides the estimation problem in two steps: a queue length estimation is made based on the fill-up time of an induc-tive loop detector located on the link in subject; and then, this “raw” measurement is fed to an Extended Kalman Filter (Anderson and Moore2012) coupled with a delay and queue length model proposed by Kimber and Hollis1979. The current solution uses the same strategy. It applies local estimations of queue lengths as measurements for a more complex model attached to an Unscented Kalman Filter (Julier et al.1995;

Julier and Uhlmann1997). By substituting the model with a CTM encompassing the whole network, it is possible to account for the influence of the dynamics of the network in the formation of the queues. Besides that, a third step is incorporated to the technique, which uses the CTM, updated with the values from the former step, to simulate future developments of the queues and functioning as a queue predictor.

The following sections will describe each component of the proposed queue predic-tion technique.