To expedite the energy transformation of the power system, the involvement of thermal power units (TPUs) in deep peak regulation (DPR) has become an effective strategy for enhancing the utilization of renewable energy. However, the optimal scheduling strategy of TPUs participating in DPR is characterized by multivariate, complex, and nonlinear mixed-integer programming problem. To address the problems of existing solution methods frequently falling into the local optimal solution and ignore the nonlinear characteristics and lead to inaccurate approximation results, an optimal scheduling model based on the multi-condition non-uniform piecewise linearization method was proposed, and the multi-condition objective function was converted into the form of non-piecewise linear equations by introducing decision variables, and the model was converted into a mixed integer linear programming (MILP) problem by using the non-uniform piecewise linearization method based on the optimal power segmentation interval of the multi-condition DPR model. Ultimately, the proposed method’s reliability and precision are confirmed through the analysis of various examples.