Backtracking, Optimizing individual trackers orientations

For "optimizing" the Backtracking on a Hill, some companies propose a control of each tracking angle, tracker-by-tracker.

This requires a sophisticated algorithm (needing machine-learning techniques), which should of course be deployed on-site for the control system of the PV plant.

We cannot understand well the principles of this development, for two main reasons:

• For backtracking (i.e. avoiding the mutual shadings, and optimize the tracker orientation), the optimization cannot be done on each tracker. The optimal position of one tracker depends on the position of the two neighbor trackers. And the same occurs for these two neighbor trackers. This means that the optimization is a delicate calculation/optimization of the position of each tracker, which are completely interdependent. The optimization can only be performed globally by a general variation of each orientation on the full PV plant.
  By the way, the previous item "Backtracking on a hill" shows that completely avoiding the mutual shadings is impossible, unless choosing a very bad orientation.

• These optimizations probably don't take the electrical losses into account (a little shade produces the loss of the full string). Now the electrical losses are the main difference in the yield between systems with and without backtracking. The previous page "Backtracking performance" shows and explains that when restricting to linear shadings, the yield of each option is very similar, due to the fact that you intercept the same "light tube".

NB: It is obviously not possible to evaluate the performance of this strategy in our simulations, as PVsyst doesn't avail of the concerned algorithm. Therefore PVsyst cannot assess the gains announced by the manufacturers.