|
|
Approximations in Stochastic Optimization and Their Applications
Mrázková, Eva ; Horová, Ivana (oponent) ; Štěpánek, Petr (oponent) ; Karpíšek, Zdeněk (vedoucí práce)
Many optimum design problems in engineering areas lead to optimization models constrained by ordinary (ODE) or partial (PDE) differential equations, and furthermore, several elements of the problems may be uncertain in practice. Three engineering problems concerning the optimization of vibrations and an optimal design of beam dimensions are considered. The uncertainty in the form of random load or random Young's modulus is involved. It is shown that two-stage stochastic programming offers a promising approach in solving such problems. Corresponding mathematical models involving ODE or PDE type constraints, uncertain parameters and multiple criteria are formulated and lead to (multi-objective) stochastic nonlinear optimization models. It is also proved for which type of problems stochastic programming approach (EO reformulation) should be used and when it is sufficient to solve simpler deterministic problem (EV reformulation). This fact has the big importance in practice in term of computational intensity of large scale problems. Computational schemes for this type of problems are proposed, including discretization methods for random elements and ODE or PDE constraints. By means of derived approximations the mathematical models are implemented and solved in GAMS. The solution quality is determined by an interval estimate of the optimality gap computed via Monte Carlo bounding technique. Parametric analysis of multi-criteria model results in efficient frontier computation. The alternatives of approximations of the model with reliability-related probabilistic terms including mixed-integer nonlinear programming and penalty reformulations are discussed. Furthermore, the progressive hedging algorithm is implemented and tested for the selected problems with respect to future possibilities of parallel computing of large engineering problems. The results show that it can be used even when the mathematical conditions for convergence are not fulfilled. Finite difference method and finite element method are compared for deterministic version of ODE constrained problem by using GAMS and ANSYS with quite comparable results.
|
|
|
Stochastic Programming Algorithms
Klimeš, Lubomír ; Mrázková, Eva (oponent) ; Popela, Pavel (vedoucí práce)
Stochastic programming and optimization are powerful tools for solving a wide variety of engineering problems including uncertainty. The progressive hedging algorithm is an effective decomposition method for solving scenario-based stochastic programmes. Due to the vertical decomposition, this algorithm can be implemented in parallel thereby the computing time and other resources could be considerably spared. The theoretical part of this master's thesis deals with mathematical and especially with stochastic programming. Further, the progressive hedging algorithm is presented and discussed in detail. In the practical part, the original parallel implementation of the progressive hedging algorithm is suggested, fruitfully discussed and tested to simple problems. Furthermore, the presented parallel implementation is used for solving the continuous casting process of steel slabs and the results are appraised.
|
|
|
Stochastic Optimization of Network Flows
Málek, Martin ; Holešovský, Jan (oponent) ; Popela, Pavel (vedoucí práce)
The master's thesis focuses on the stochastic optimization in network flow problems. The theoretical part covers three topics - the graph theory, the optimization and the progressive hedging algorithm. Within the optimization the main part is devoted to the stochastic programming and the two-stage programming. The progressive hedging algorithm includes also the scenario aggregation and the modification of the general algorithm to two-stage problems. The practical part deals with models using real-world data of collection of municipal waste within the Czech Republic, which were provided by the Institute of Process Engineering.
|
|
|
Spatial Decomposition for Differential Equation Constrained Stochastic Programs
Šabartová, Zuzana ; Mrázková, Eva (oponent) ; Popela, Pavel (vedoucí práce)
Wide variety of optimum design problems in engineering leads to optimization models constrained by ordinary or partial differential equations (ODE or PDE). Numerical methods based on discretising domain are required to obtain a non-differential numerical description of the differential parts of constraints because the analytical solutions can be found only for simple problems. We chose the finite element method. The real problems are often large-scale and exceed computational capacity. Hence, we employ the progressive hedging algorithm (PHA) - an efficient scenario decomposition method for solving scenario-based stochastic programs, which can be implemented in parallel to reduce the computing time. A modified PHA was used for an original concept of spatial decomposition based on the mesh created for approximation of differential equation constraints. The algorithm consists of a few main steps: solve our problem with a raw discretization, decompose it into overlapping parts of the domain, and solve it again iteratively by the PHA with a finer discretization - using values from the raw discretization as boundary conditions until a given accuracy is reached. The spatial decomposition is applied to a basic test problem from the civil engineering area: design of beam cross section dimensions. The algorithms are implemented in GAMS software and finally results are evaluated with respect to a computational complexity and a length of overlap.
|
|
|
Stochastic Optimization of Network Flows
Málek, Martin ; Holešovský, Jan (oponent) ; Popela, Pavel (vedoucí práce)
The master's thesis focuses on the stochastic optimization in network flow problems. The theoretical part covers three topics - the graph theory, the optimization and the progressive hedging algorithm. Within the optimization the main part is devoted to the stochastic programming and the two-stage programming. The progressive hedging algorithm includes also the scenario aggregation and the modification of the general algorithm to two-stage problems. The practical part deals with models using real-world data of collection of municipal waste within the Czech Republic, which were provided by the Institute of Process Engineering.
|
|
|
Approximations in Stochastic Optimization and Their Applications
Mrázková, Eva ; Horová, Ivana (oponent) ; Štěpánek, Petr (oponent) ; Karpíšek, Zdeněk (vedoucí práce)
Many optimum design problems in engineering areas lead to optimization models constrained by ordinary (ODE) or partial (PDE) differential equations, and furthermore, several elements of the problems may be uncertain in practice. Three engineering problems concerning the optimization of vibrations and an optimal design of beam dimensions are considered. The uncertainty in the form of random load or random Young's modulus is involved. It is shown that two-stage stochastic programming offers a promising approach in solving such problems. Corresponding mathematical models involving ODE or PDE type constraints, uncertain parameters and multiple criteria are formulated and lead to (multi-objective) stochastic nonlinear optimization models. It is also proved for which type of problems stochastic programming approach (EO reformulation) should be used and when it is sufficient to solve simpler deterministic problem (EV reformulation). This fact has the big importance in practice in term of computational intensity of large scale problems. Computational schemes for this type of problems are proposed, including discretization methods for random elements and ODE or PDE constraints. By means of derived approximations the mathematical models are implemented and solved in GAMS. The solution quality is determined by an interval estimate of the optimality gap computed via Monte Carlo bounding technique. Parametric analysis of multi-criteria model results in efficient frontier computation. The alternatives of approximations of the model with reliability-related probabilistic terms including mixed-integer nonlinear programming and penalty reformulations are discussed. Furthermore, the progressive hedging algorithm is implemented and tested for the selected problems with respect to future possibilities of parallel computing of large engineering problems. The results show that it can be used even when the mathematical conditions for convergence are not fulfilled. Finite difference method and finite element method are compared for deterministic version of ODE constrained problem by using GAMS and ANSYS with quite comparable results.
|
|
|
Spatial Decomposition for Differential Equation Constrained Stochastic Programs
Šabartová, Zuzana ; Mrázková, Eva (oponent) ; Popela, Pavel (vedoucí práce)
Wide variety of optimum design problems in engineering leads to optimization models constrained by ordinary or partial differential equations (ODE or PDE). Numerical methods based on discretising domain are required to obtain a non-differential numerical description of the differential parts of constraints because the analytical solutions can be found only for simple problems. We chose the finite element method. The real problems are often large-scale and exceed computational capacity. Hence, we employ the progressive hedging algorithm (PHA) - an efficient scenario decomposition method for solving scenario-based stochastic programs, which can be implemented in parallel to reduce the computing time. A modified PHA was used for an original concept of spatial decomposition based on the mesh created for approximation of differential equation constraints. The algorithm consists of a few main steps: solve our problem with a raw discretization, decompose it into overlapping parts of the domain, and solve it again iteratively by the PHA with a finer discretization - using values from the raw discretization as boundary conditions until a given accuracy is reached. The spatial decomposition is applied to a basic test problem from the civil engineering area: design of beam cross section dimensions. The algorithms are implemented in GAMS software and finally results are evaluated with respect to a computational complexity and a length of overlap.
|
|
|
Stochastic Programming Algorithms
Klimeš, Lubomír ; Mrázková, Eva (oponent) ; Popela, Pavel (vedoucí práce)
Stochastic programming and optimization are powerful tools for solving a wide variety of engineering problems including uncertainty. The progressive hedging algorithm is an effective decomposition method for solving scenario-based stochastic programmes. Due to the vertical decomposition, this algorithm can be implemented in parallel thereby the computing time and other resources could be considerably spared. The theoretical part of this master's thesis deals with mathematical and especially with stochastic programming. Further, the progressive hedging algorithm is presented and discussed in detail. In the practical part, the original parallel implementation of the progressive hedging algorithm is suggested, fruitfully discussed and tested to simple problems. Furthermore, the presented parallel implementation is used for solving the continuous casting process of steel slabs and the results are appraised.
|
host ::
RSS.