This paper introduces the demand-driven assembly line rebalancing problem (DDALRP) and proposes a non-linear, multi-objective, combinatorial optimization model to solve it. A DDALRP arises whenever the production output of the assembly line (AL) must be continuously readjusted along a planning horizon in order to satisfy as much as possible a given demand forecast; thus, dealing not with a one-time rebalance, but with a multi-period rebalance, fact that exponentially increases the complexity and combinatorial nature of the problem. Adapting or regulating the production output of the AL to some demand forecast or production plan is a relatively new idea in the assembly line balancing (ALB) / rebalancing (ALR) literature; and we address this problem by reallocating workers to stations, taking into consideration their learning and forgetting (L&F) curves. Our proposed model was solved by implementing a genetic algorithm (GA) in 162 cases (three problem instances under 54 scenarios each), obtaining useful insights about the dynamic of worker reallocation under different scenarios: optimistic, most-likely, pessimistic L&F coefficients; experienced and inexperienced workers; and different demand scenarios.