Construction of Statistical Models with High Estimation Accuracy for Profit Functions

Abstract

In this paper, we study the problem of finding a control input plan that maximizes a profit function based on a statistical model. We employ covariate shift adaptation so that the statistical model has a high prediction ability at important covariate values. The degree of importance is determined by the profit function. To determine a control input plan based on the statistical model, we develop an iterative method that updates the statistical model and the control input plan alternately. We illustrate the effectiveness of the iterative alternate optimization method by a charge/discharge plan problem of a battery.