Optimizing prior prediction accuracy for disturbances in guaranteed control problems

We consider and formalize the problem of finding optimal accuracy for prior prediction of disturbances in guaranteed planning problems under the assumption that there exist ready algorithms for processing observations able to increase prediction quality at some significant added costs. We analyze the general qualitative scheme of this problem and find analytic dependencies of optimal prediction quality for an illustrative problem of delivery planning. With the example of this problem, we show how useful it is to establish connections between different components in the disturbance vector. We propose a scheme of step-by-step increase in prediction accuracy that does not require prior knowledge of the prediction costs function.