Circulating fluidized bed (CFB) is becoming the main coal combustion technique all over the world for its high combustion efficiency, low pollution, fuel flexibility and good load-following capability, but its combustion process characteristic is highly non-linear, time-varying, large time delay and multi-variable strong coupling. The conventional control proposal divides the bed temperature control system and main steam pressure control system into two separate PID single-loops without consideration of inter-circuit coupling, which results in low automatism operation rate and often manual operation required. To solve this problem, a multi-variable adaptive neural network decoupling controller with PID structure is introduced based on the principle of NN control and decoupling compensation. The decoupling and control capability of PID NN decoupling controller are from the NN cross-tie structure and the nonlinear mapping properties, as well as the PID style processing of the hidden layer node. The weights of the NN are learned and optimized by the particle swarm optimization (PSO) algorithm, which is known for its searching ability in the total parameter space concurrently and efficiently. These NN weights will not only eliminate the coupling relations between circuits, but also strengthen the PID-NN controller's adaptability. This multivariable decoupling control method is used to control the two-dimensional transfer function matrix of the bed temperature and main steam pressure system in the CFB combustion system. Simulation results show that the new control strategy could overcome nonlinear and strong coupling features of CFB combustion system in a wide range and is expected to have great potential for engineering application.