This paper proposes a methodology to solve generation expansion equilibrium problems by using a predictive model to represent the equilibrium in a simplified network constrained electricity market. The investment problem for each generation company (Genco) is a bi-level problem with the investment decision made in the upper level and market clearing condition in the lower level, which traditionally is represented as a Mathematical Program with Equilibrium Constraint (MPEC). The predictive model is trained for estimating the system-wide revenues for each technology type across energy, ancillary services and capacity markets given the amount of technology-specific installed capacity on the grid. The profit maximization investment problem for each Genco is solved using a global search algorithm, which uses the predictive model to evaluate the objective function. To solve for the strategic equilibrium, each Genco's problem is plugged into a diagonalization algorithm that is generally used in multi-leader, single-follower bi-level problems. The methodology presented here enables significant computational improvements while still capturing the desired market characteristics and dynamics of traditional equilibrium modeling approaches
