The invention provides a method for controlling the combustion process of a circulating fluidized bed boiler on the basis of multivariable generalized predictive control optimization. The method comprises the steps that 1, the design parameters of controllers for the combustion process of the circulating fluidized bed boiler are initialized; 2, a combustion process model is built through identification of a multivariable gradual reducing memory recursive least-square method according to collected process variable data of the combustion process of the circulating fluidized bed boiler; 3, the parameters of the proportion-integration-differentiation (PID1) controller, the PID2 controller and the PID3 controller are subjected to optimal setting according to multivariable generalized predictive control characteristics, and setting results are decoupled; 4, according to the obtained parameters of the controllers, the coal feed amount, the primary air volume and the secondary air volume are controlled respectively, and therefore the bed temperature, the main steam pressure and the oxygen content of smoke of a hearth outlet are controlled; and 5, the next moment starts, and the step 2 to the step 5 are repeated. According to the method, the control precision is high, the tracking speed is high, overshooting is avoided, steady state errors are small, the anti-coupling capacity is high, and the control process is smooth; the method is a multivariable control technology, which is simple in mode and convenient to implement, for the combustion process of the circulating fluidized bed boiler.