Thesis defence A. Jamshidnejad: urban mobility

22 June 2017 10:00 - Location: Aula, TU Delft - By: Webredactie

Efficient predictive model-based and fuzzy control for green urban mobility. Promotor: J. Hellendoorn (3mE).

Efficient PredictiveModel-Based and Fuzzy Control for Green UrbanMobility
In this thesis, we develop efficient predictive model-based control approaches, including model-predictive control (MPC) and model-based fuzzy control, for application in urban traffic networks with the aim of reducing a combination of the total time spent by the vehicles within the network and the total emissions. The thesis includes three main parts, where in the first part the main focus is on accurate approaches for estimating the macroscopic traffic variables, such as the temporal-spatial averages, from a microscopic point-of-view. The second part includes efficient approaches for solving the optimization problem of the nonlinear MPC controller. The third and last part of the thesis proposes an adaptive and predictive model-based type-2 fuzzy control scheme that can be implemented within a multiagent control architecture.

Flow and density are mainly used to characterize partly the state of physical systems with moving particles. Flow and density are macroscopic concepts, i.e., they involve average values. For dynamic systems that include moving particles, such as for traffic networks, these averages may be defined in three different ways: temporal, spatial, and temporal-spatial. Computation of the first two averages is straightforward, but for the third average only a general formulation is suggested by Edie (1963), while details regarding computation of this general formula are missing in literature. Since flow and density play a prominent role in model-based analysis and control of long traffic roads, in the first part of the thesis, we focus on microscopic approaches for accurate estimation of these variables in the temporal spatial sense. The proposed approaches can be applied to any dynamic physical system with moving particles, in particular traffic networks, which are the main concern of this thesis.

The second part of the thesis is focused on developing efficient and fastmodel-predictive control approaches for systems with a highly nonlinear behavior (such as traffic networks), gradient-based optimization is an efficient and fast method for finding local optima of a nonlinear function. For many nonconvex functions, this approach can still be efficiently applied considering multiple initial starting points for searching. We apply this approach to solve the optimization problem of the MPC controller. In this context, we discuss two cases that may occur in solving a nonlinear MPC optimization problem: smooth case and nonsmooth case, where the first one can be dealt with via gradient-based methods. We then develop general smoothening approaches to readjust nonsmooth optimization problems into smooth ones that can be solved by a gradient-based optimization method. The resulting control system is implemented in an urban traffic network with the aim of finding a balanced trade-off between prevention/reduction of traffic congestion and decreasing the level of emitted pollutants. Additionally, to predict the future evolution of the states of the traffic network in a reliable way and within a reasonable time span, we develop a general framework to integrate and interface macroscopic traffic flow models and microscopic emission models, which results in a computationally efficient and accuratemesoscopic traffic flow and emission model that can be used as predictionmodel forMPC.

In the third and last part of the thesis, we combine predictive control methods with model-based fuzzy control approaches to develop a two-layer adaptive fuzzy control system that can potentially be used in a coordinativemulti-agent architecture, in particular for controlling processes with time-delayed input and states. We first extend the affine formulation of the Takagi-Sugeno model-based fuzzy approach to a nonlinear one, and propose a general nonlinear type-2 fuzzy formulation for modeling time-delayed systems. The proposed fuzzy rules are constructed from type-2 fuzzy sets, which can deal with uncertainties and ambiguities regarding the available information for themodeled system. This new formulation is applied to both the controller and the predictionmodel of the system. We decompose the overall minimization problem into local ones using primal decomposition, in order to find the optimal value of the local performance indices and to steer the coordination of the agents in such a way that the realized value of each local performance index never exceeds the optimal one.

In summary, themain contributions of this thesis are

  • We introduce accurate microscopic approaches for estimating the temporal-spatial macroscopic traffic variables.
  • We develop efficient methods for solving the nonlinear optimization problem of the MPC controller by smoothening the onsmooth optimization problemand via implementing a gradient-based optimization approach.
  • We introduce a generalmesoscopic framework for integratingmacroscopic traffic flow models with microscopic emission models. The resulting model can provide fast and accurate estimates of the future emissions and the future total time spent by the vehicles.
  • We propose an adaptive nonlinear fuzzy and model-based predictive control scheme that can be used within a multi-agent control architecture, in particular, for processes that involve time-delayed input and/or states.

More information?
For access to theses by the PhD students you can have a look in TU Delft Repository, the digital storage of publications of TU Delft. Theses will be available within a few weeks after the actual thesis defence.