A novel unit maintenance scheduling (UMS) problem formulation for a generation producer is presented, to maximise its benefit while thoroughly considering the risk associated with unexpected unit failures. First, the unit failure is characterised by a more practical bathtub-shaped failure behaviour from the modified superposed power law process. Its parameters are estimated from the historical data by solving a nonlinear least-squares fitting problem via Gauss-Newton iteration method. On the basis of the unit failure analysis, the new UMS formulation is solved by a combination of linear programming and genetic algorithms (GAs), and its impact on the producer's benefit is analysed in detail, including expected profit of selling energy, expected renewal cost of damaged components and maintenance cost. Compared with the current models, the proposed UMS model takes into consideration the influences of market factors as well as unexpected unit failures to strike the right balance between profits and costs related with potential unit failures. Numerical examples on a four-unit producer are utilised to demonstrate the effectiveness of the proposed scheme.