In many engineering problems, an optimal solution must be derived by considering reality situations that reflect actual measured data. In the optimization problem, the objective function is modeled through a formula. Such modeling is difficult in reality situations that are affected by multiple environments, such as temperature and humidity. Therefore, this paper models the objective function as a data-driven approach under deep-learning-based nonlinear regression analysis, and we then propose the constraint optimization problem in a specific environment (trained DNN objective optimization). We define the solution to this problem as a controllable local-optimal solution (CLS) and propose an environment parameter fixed algorithm (EPFA) to derive the CLS. This changes constraint optimization into an unconstrained optimization problem by pinning down environment parameters to reflect constraints in a particular environment. After that, the proposed approach is combined with conventional Gradient Descent and algorithms such as Adagrad to derive the CLS. The situation is explained through an example of an optimal course model created for use in this study. In addition, we verify that the CLS can be derived through experiment by using an optimal course dataset and a Boston house price dataset.