Workflow management systems support the automatization of business processes so that efficiency and productivity are enhanced. An important feature which improves the process quality, is the ability to monitor the compliance to specified constraints. A very frequently applied class of constraints comprises the integration of time and time restrictions, for instance deadlines, into a process.
The violation of such a constraint will increase the cost of a process as it often requires expensive escalation-actions to adjust the situation or even entails the payment of penalty fees.
\bigskip oindent Therefore it is reasonable to integrate mechanisms that allow the prediction and proactive avoidance of time constraint violations. The basic idea is to forecast the execution behavior and task durations of processes by calculating valid execution intervals for each task. This provides the means of monitoring the execution of tasks within their boundaries and to initiate evasive actions if they are prone to violating these thresholds. Although the immense optimization potential of such an idea is rather obvious, existing research approaches still did not receive the attention they deserve in commercial workflow systems. The main reason for that is definitely to be found in uncertainties during process execution.
They are, on the one hand, caused by conditional process structures which inhibit exact prediction of upcoming tasks, and, on the other hand, caused by the impossibility of forecasting the exact duration of a task that varies with each execution. Additionally existing research approaches are restricted to very basic workflow control flow elements, which inhibits application in commercial workflow systems.
\bigskip oindent This thesis describes a probabilistic time management approach, which utilizes empirical knowledge that has been extracted from workflow logs. Forecasts are based on probabilistic timed process graphs, which represents temporal information, like the above-mentioned intervals, in the form of time histograms. These histograms are used to forecast the likelihood of future constraint violations, upcoming activity assignments, and expected remaining execution times. The thesis also shows how to deal with advanced control flow structures with a special focus on arbitrary cycles. Furthermore it describes how to apply proactive time management, such that future deadline violations can be avoided, along with the simulation of several scenarios to demonstrate the benefits of a probabilistic time management approach.