Ole-Erik Vestøl Endrerud

09 February 2017

Now most of us agree that weather is the primary source of uncertainty in an offshore construction project and a significant restriction on maintenance activities. And to properly understand that risk and its implication on performance, we have three different approaches to analyze weather; historic, vary start year, and randomised. The aim of this post is to explain that using a simulation model that simulates actual operations provide a much more accurate and realistic weather downtime assessment. And, the data used to generate the weather downtime statistics from a traditional statistical approach is the same as we use in our Shoresim applications to simulate and generate weather downtime statistics, among many other performance metrics.

Based on these time series, such as the one above, you can generate weather window statistics and subsequently traditional weather downtime statistics. The weather window statistics is a histogram of weather window lengths, for a specified weather state. A weather state consists of one or more weather parameters. E.g. the pair (wind speed, wave height) is a weather state. So, when generating weather window statistics, we count how many weather windows there are for a weather state with say 1 hour, 2 hours, 3 hours, … 30 hours, etc. lengths in a weather data time series.

The traditional statistical approach to calculate weather downtime is a deterministic and analytic one. Meaning the operations interdependencies, the sequence of execution, timing and where the operations are done isn’t considered by the statistical method. In a statistical approach, the weather downtime statistic is derived by looking at how many weather windows there are in a month that satisfies a required weather window length. In other words, a statistical approach can calculate how many weather windows that exist in a weather data time series for a particular activity`s duration and weather state restriction.

With a simulation approach the installation campaign or maintenance operations are simulated operation by operation in the time domain (in minutes and hours), in the correct sequence with all interdependencies, timing and location of operations, which vessel does it and what support activities are required (e.g. the need for several vessels). This means that it isn’t just important to know if there are any weather windows to execute an operation during a month, but will there be weather windows to do several activities in sequence with the right timing, taking into account all activities that need to be carried out. So a much more realistic approach.

How does Shoresim treat weather

On the top, we said our Shoresim apps could use weather in three different ways. We’ll get to that, but first, let’s explain how our Shoresim apps uses weather time series in simulations. Then, because our Shoresim apps MAINTSYS and SIMSTALL are simulation tools that simulate the actual sequence and timing of operations in the time domain, they require weather input in the time domain; and not statistical input such as weather windows, weather downtime, or what some also call workability.

The weather data time series (such as the one above) is used by our weather checking algorithm to check if the weather criteria for an activity are met for the required weather window length to carry out that activity.

We’ll illustrate this with an example. Let’s use a construction project with 100 wind

turbine installations in the SIMSTALL app as a case. For all these 100 wind turbine installations the heavy lift vessel will go through a work process to assemble all parts. This work process has three steps in this example: lift the tower in place, lift the nacelle in place and lastly the three blades. Each of these steps has a duration of 7 hours but require a 9-hour weather window. These three activities have different weather criteria, as you can see below.

Our weather checking algorithm will run for each of these activities to check if they are allowed to start. What happens then during the simulation is that the installation vessel will transit to the wind turbine location and jack up (with their corresponding weather criteria), and check if weather permit the Tower installation activity to start. It checks that wind speed is less than 14 meters per second at a 100-meter height for the next 9 hours. If these criteria are met, this activity starts. When the tower installation is done a new weather window check for a window of 9 hours with wind speed below 14 m/s at 100-meter height for the nacelle installation. And at last the same for blades with a stricter wind speed limitation. If the tower installation were delayed because of wind speeds, that would cause a delay that might cause the nacelle installation to miss a weather window, resulting in more weather downtime. This interdependence between activities isn’t caught with the statistical approach.

To make it short estimating weather downtime with a simulation approach is much closer to reality and the actual weather downtime, because it takes into account the interdependence, sequence, and timing of activities.

Three ways to use weather time series: raw, change start year, and randomize

The second option is to increment the start year in your weather data time series. This approach is to use your imported weather data directly in your simulation runs, but for the first simulation run the start year is 1980, for the second run it is 1981, etc. That way you will get output and weather downtime based on starting in any of the years in your weather data set. A good way to look at the weather risk in a project based on actual measured weather.

Lastly, you can create synthetic weather time series based on your own imported weather time series, if you choose to do so. Our Shoresim apps will then preprocesses your weather data into weather states and uses a Markov model to generate a new weather data time series that have the same characteristics (within a 0.5-3% margin of error) regarding weather window statistics. What the Markov model does is to calculate the probability of going from a weather state (e.g. 1.5 m wave height and 10 m/s wind) to all other weather states that have appeared after that one in your weather data set. Based on these probabilities, you can generate a new string of weather states by drawing randomly generated numbers between 0 and 1 for each new state and picking the next state with the closest probability. There is a lot of literature out there on this method, see for example this one from Hagen et al., 2013: "A Multivariate Markov Weather Model for O&M Simulation of Offshore Wind Parks.” Creating synthetic weather data time series is very useful when your weather data time series is incomplete, or you want to run more simulation runs than you have years of data in your own weather data time series. And the accuracy of your newly generated time series improves with the amount of raw data you import of course.

To conclude, using a simulation model that simulates actual operations provide a much more accurate and realistic weather downtime assessment. And the data used to generate the weather downtime statistics from a traditional statistical approach is the same as we use in our Shoresim apps to simulate and generate weather downtime, among many other performance metrics.