Aalto / ICS / Research / CAC / OLP |
Main page Background and Objectives Production Planning Online Algorithms RobuTu Results Links |
Results of the OLP projectThe project's scientific results are presented in 3 conference papers and 4 journal articles (one of them an invited full version of one of the conference papers). The main contributions per each publication are as follows: Conference paper [1] considers short-term production planning in a multi-item factory with capacity constraints. The factory produces items that are shipped to a warehouse. The warehouse has an uncertain daily demand for each of the item types. Some items have a nearly constant demand, while others are purchased occasionally. The goal is to minimise the number of back-ordered items, while keeping the stock size reasonable. To achieve this, target inventory level- and production priority rule-based solutions are compared. In the examples initial stock sizes range from the target level to a situation where there are no stocks. Simulations show that, in the present problem, priority rules that take into account the whole demand distribution are better than average-based ones. Journal article [2] analyses, using the so called "competitive ratio" performance metric for online algorithms, the single commodity Economic Lot-Sizing (ELS) problem with lookahead. In this make-to-order inventory model, orders expressing demand within a given near future lookahead window are processed together, without any knowledge about demands beyond the lookahead time. It is shown that competitive ratio analysis can be applied to this problem only if the orders have a minimum unit size. Under this condition, lookahead-dependent lower bounds on the competitive ratio of any online ELS algorithm are derived; these bounds can be viewed as reflecting the value of future information. Furthermore, a new lookahead-sensitive ELS algorithm scheme GREEDYSPLIT is presented and shown to be asymptotically competitive-ratio optimal. Conference paper [3] and its invited journal version [5] analyse an online backup scheme for an independent data-collecting device. The total data collecting period is unknown and memory on the device is limited; nevertheless the data collected on the device should represent the whole time period as uniformly as possible. The paper(s) present a performance measure for this problem setting and give some simple well-performing algorithms. One of the algorithms is shown to be asymptotically optimal as the number of backup slots available to the device increases. Conference paper [4] discusses a general framework of "constraint satisfaction games" that models many adversarial situations involving sequential decisions. (The framework applies also in the worst-case analysis of online algorithms, where the "adversary" corresponds to the algorithm's environment behaving in the worst possible way.) A constraint satisfaction game comprises two players addressing a system of Boolean constraints, in which the players take turns in selecting one of the available variables and setting it to "true" or "false", with the goal of maximising (for Player I) or minimising (for Player II) the number of satisfied constraints. Unlike in standard QBF-type variable assignment games, the framework in [4] imposes no order in which the variables are to be played. This makes the game setup more natural, but also more challenging to control. The paper provides polynomial-time, constant-factor approximation strategies for Player I when the constraints are parity functions or threshold functions with a threshold that is small compared to the arity of the constraints. The results in the paper also establish that the problem of determining if Player I can satisfy all constraints is PSPACE-complete even in this unordered setting, and when the constraints are disjunctions of at most 6 literals. Journal article [6] (submitted) addresses the problem of online data aggregation in communication networks, whereby messages arriving at the nodes of a network are to be forwarded to its root, with the option of being merged with other messages along their transmission path. The goal is to minimise, for a given message sequence, the total sum of the transmission link costs and the delay penalties the messages accrue while waiting for merges. The article presents for this problem an O(log C_MST)-competitive algorithm that covers networks of bounded treewidth and so-called outerterminal planar graphs, where C_MST is the cost of a minimum spanning tree in the network. These are the first such results for non-treelike networks. The general algorithm scheme is based on a relation between the data aggregation problem and the Prize-Collecting Steiner Tree (PCST) problem, and it extends to any other family of networks where the PCST problem can be solved efficiently. An auxiliary result of some independent interest is the exact polynomial-time PCST algorithm presented in [6] for outerterminal graphs. Journal article [7] (submitted) compares three production line rescheduling policies: an unrestricted one where previously scheduled jobs can be moved freely, one where jobs can only be moved forward in the schedule, and one where already scheduled jobs cannot be moved at all. The comparison is performed by considering the minimisation of tardiness. While unrestricted rescheduling should generally give the best solution, moving jobs only forward can be more practical as e.g. in general production, material orders can be delayed but seldom advanced. The results in the paper show that rescheduling by moving jobs only forward gives a tardiness that is similar to unrestricted scheduling when small or special cases are considered. In addition, numerical experiments show that there are only small differences in larger and more general problems as well. Further, the researchers took part in the ROADEF/EURO Challenge 2010 called "A large-scale energy management problem with varied constraints". The problem consisted of number of power plants and the task was to select their maintenance periods and production levels such that they satisfy a set of constraints and optimize costs over given scenarios. The team Ahlroth, Tokola, Schumacher obtained a honourable third place in the student category. |