With the diversification of electricity price structures, an increasing number of power utilities have incorporated demand charges into their tariff structures. Individual households equipped with photovoltaic (PV) and energy storage systems can adapt to this trend by adopting appropriate energy management strategies. Due to the high uncertainty in photovoltaic and load generation for household users, it is often challenging for them to make the most advantageous choice among the diverse electricity tariffs. This paper proposes a rolling prediction method based on Long Short-Term Memory (LSTM) networks for monthly peak power demand, taking into account historical peak power and applying corrective measures within the same month. Additionally, a profit evaluation method for electricity tariff schemes considering forecast uncertainties is presented. The predictive capabilities of load and PV power are characterized using kernel density estimation, and a large number of scenarios are generated using Monte Carlo simulation. A probabilistic economic evaluation is conducted for different tariff schemes, enabling the optimal selection of electricity tariffs. To validate the effectiveness of the proposed methods, analysis is performed using household data from Arizona, and the results demonstrate that the proposed methods can reduce electricity expenses and help households choose the correct electricity tariff scheme.