A time series is a sequence of data assigned to specifi c moments in time. Most statistical models are static, such as regression analysis: The defi ning set of parameters has fixed values, and therelationship between the explanatory variables and the response is viewed as constant. This is a perfectly valid assumption for many applications, but when working with time series data, it isimportant to acknowledge that such relationships may be altered through the passage of time. As opposed to classical time series models, which are static, the Bayesian approach is based ondynamic learning, and allows for varying parameters: As new information is available sequentially, beliefs regarding the parameters expressed through a probability distribution, are updatedusing Bayes' theorem. Intuitively, recent data are more valuable than older data when making inference on current events. This information loss is recognized when using dynamic models, while for classical time series, all information is weighted equally as the model parameters are static. In addition to its appealing dynamic properties, inference and interpretation of Bayesian time series results are intuitive and straightforward (as for Bayesian statistics in general). As the complete Bayesian time series framework is based on one single theorem, Bayes' theorem, the theory is simplifi ed and unifi ed. The Bayesian paradigm is also particularly suitable forprediction, taking into account all parameter uncertainties, as well as model uncertainty.
In this thesis our main aim is to apply Bayesian dynamic models, as thoroughly presented in West & Harrison (1997), to a univariate strain time series from the Göta River Bridge, and find a model that provides good short and long term predictions. The Göta River Bridge connects Gothenburg's mainland to the island Hissingen. The bridge is over 70 years old, and the steel beams are of relatively poor and varying quality. During the 90s, several minor cracks and fatiguedamages were discovered in the bridge structure. The bridge went through major repair, but cracks due to fatigue may occur again, and lead to collapse of the steel girders. Swedish traffic authorities have decided to keep the bridge in service for another 10 to 15 years, but in order to increase safety, the condition of the bridge must be monitored continuously. The Norwegian research center NGI was assigned the job of providing a surveillance system, and has installedover 5km of fi ber optics to monitor for increasing deformations. The system provides real-time strain data every other hour, for over 50000 points along the bridge girders.