The mixed integer quadratic programming（MIQP） algorithm for solving unit commitment problems has global optimization capability. For a large-scale optimization problem， however， its computation speed and accuracy will be affected. To deal with this problem， an improved mixed integer quadratic programming algorithm is presented. The crucial idea of the algorithm is the introduction of two kinds of improvement strategies of relaxation and decoupling. First， the lower bound space of unit commitment is obtained by solving a quadratic integer programming model with slacking integer variables. Then， the upper bound space of unit commitment is obtained using the Lagrangian decoupling algorithm. Finally， in the optimizing space determined by the upper and lower bounds， unit commitment is re-optimized with the MIQP algorithm. Different simulation results show that the improved MIQP algorithm is fast and effective and capable of reducing the complexity of optimization， especially the computational time.