We consider the discrete time optimal control problem for a battery in the electricity intraday market, which is linked to a photovoltaic power station. We discuss recent Regression Monte Carlo (RMC) methods to solve the stochastic backward dynamic programming equations arising in the problem. We propose adaptive model selection methods, which reduces the complexity of the control optimization while improving the accuracy for small simulation budgets. Furthermore, we introduce a control map approximation, which further reduces complexity of the algorithm with negligible loss of accuracy and allow for an efficient nested simulations approach.