A technique is disclosed for controlling a material blending process for blending a plurality of raw material feeds using model predictive control in order to produce a blended product conforming to one or more quality standards. The quality standards may be based upon control variables derived from relationships and/or ratios of certain materials in the composition of the raw material feeds (e.g., moduli). The determined control variables are compared to desired set points and, if a deviation from the desired set points is detected, the proportions of the raw material feeds are adjusted to align the control variables with the desired set points. The technique may further take material cost factors into account, wherein the prediction model determines a proper proportioning of the raw material feeds such that the blended product conforms to the desired specifications and is produced for the lowest cost.