This paper proposes a support vector regression (SVR)-based model predictive control (MPC) for the volt-var optimization (VVO) of electrical distribution systems. First, measurement data from a few days of operation of a distribution system, gathered using advanced metering infrastructure (AMI), are used to train an SVR model of the system. The trained model is then employed by the MPC in a closed-loop control scheme to control capacitor banks and tap changers of the distribution system so that the power loss is minimized, and voltage profiles are maintained within a specific range. In contrast to the many existing VVO methods, the proposed scheme does not require any circuit-based simulations for its operation, nor does it assume that the distribution system is radial. The simulation results of applying the proposed SVR-based MPC to IEEE123 bus test feeder proves that despite its measurement-based feature, the proposed approach is capable of providing close to optimal solutions to the VVO problem. The simulation results also suggest a satisfactory outcome of the proposed approach in controlling meshed grids or in the presence of distributed energy resources (DERs).INDEX TERMS Model predictive control, parallel optimization, power distribution, support vector regression, volt-var optimization.