The increasing demand for electricity offers many opportunities for renewable energy production, of which one alternative is tidal stream energy. Several feasibility studies have shown that the global tidal stream energy potential can contribute significantly to producing renewable energy. This tidal energy can mostly be produced at the 'tidal hotspots', where the kinetic energy density is very high due to fast flowing tidal currents. However, the tidal technology is not yet cost competitive in comparison with other renewables, such as photovoltaic and wind energy, which is why further cost reductions and efficiency improvements are to be achieved. Interviews with existing tidal system developers provided insight in the cost breakdown and showed that maintenance accounts for a significant share of the total project costs. This is due to the harsh environmental conditions that impose a large uncertainty, which increase the complexity of selecting an optimal maintenance policy. Damen Shipyards has shown interest in entering the tidal industry and is exploring the cost reduction possibilities by developing their own tidal system. This thesis contributes to Damen Shipyards' research by performing a time series analysis of a tidal hotspot to identify and model the multivariate dependence of the governing environmental phenomena. A probabilistic decision support tool is developed for selecting the optimal maintenance policy. The decision support tool primarily determines when and to what extent corrective maintenance should be performed. The corresponding overall maintenance costs are also calculated and secondary information regarding the activity duration is given. By means of the probabilistic approach, which captures the weather window uncertainty due to the environmental randomness, the results can be interpreted by the user based on the desired confidence level. In this research the weather window uncertainty is implemented by simulating a large number of random, but statistically identical environmental time series, which are based on available measurement data of the tidal field at EMEC, located at the Orkney Islands in the United Kingdom. The multivariate dependence between the significant wave height, wave peak period, wind velocity and current velocity is identified in the measurement set and fully represented in the generated time series by means of a pair-copula construction simulation. The necessity for having time independence cannot be met in the original dataset, which is why a new simulation approach is developed. This method consists of a sequential simulation of pair-copula constructions to include both the time dependence and multivariate dependence in the synthetic time series. Simulation of the set of synthetic time series showed to be more effective for describing uncertainty with respect to exclusively using the original dataset, due to the possibility of including more environmental realizations. The tidal array is represented as a semi-Markov decision process, which captures all costs and transition processes related to the deterioration and maintenance decisions. A policy optimization algorithm can then be used to find the optimal set of decisions and the corresponding maintenance cost rate which includes both the direct and indirect maintenance costs. The novel tidal system design of Damen Shipyards is then plugged into the decision support tool to determine the optimal maintenance policy and maintenance costs. The effect of different levels of detail for representing the tidal system have been compared and the benefits in terms of cost reductions of using this decision support tool with respect to less advanced approaches have been highlighted. Furthermore, multiple scenarios have been elaborated to identify the sensitivities in the cases of accounting for unreliability in the failure rates, varying the number of platforms in the array and including the economic fluctuations of the maintenance vessel day rates.