Process controlling method with merged two-control loops
The architecture of the controller of the present invention is mainly composed of three portions, which are a model predictive controller, a local controller (also called stably assistant controller), and a coordinator. This modified model predictive control structure merges two control loops by a coordinator. The control mode can be switched into either of them or provide a combined control action by adjustments of the coordinator. When the system moves out the range of the database, the coordinator will automatically detect the fault and switch to the stably assistant controller to provide the basic control quality to ensure the safety.
1. A process controlling method with merged two-control loops comprising the following steps:
i) inputting a value of y k of a controlled variable of a process at a sampling time point of k to a coordinator, so that the coordinator outputs a weighting, ψ, wherein 0≦ψ≦1, and k is 0 or a positive integer;
ii) inputting said y k to a model predictive controller (MPC), so that a MPC output of a manipulated variable, u MPC , is obtained from said MPC;
iii) inputting said y k to a stably assistant controller (SAC), so that a SAC output of said manipulated variable, u SAC , is obtained from said SAC;
iv) calculating a combination output of said manipulated variable, u, according to the following formula:
u=ψ·u MPC +(1−ψ)* u SAC
wherein ψ, u MPC , and U SAC are defined as above; and
v) adjusting said process by using said combination output; wherein
in step i), ψ is determined according to a calculation of a regional knowledge index function, said regional knowledge index function is established by evaluating performance of the MPC alone in controlling the process with test data so that said regional knowledge index function reflects whether the value of y k falls in a well trained region, an unfamiliar region or an intermediate region between these two regions, wherein ψ=1, when the value of y k falls in the well trained region; ψ=0, when the value of y k falls in the unfamiliar region; and 0<ψ<1, when the value of y k falls in the intermediate region, wherein ψ approaches 0, when the value of y k is getting closer to the unfamiliar region, and ψ approaches 1, when the value of y k is getting closer to the well trained region;
in step v), said adjusting will cause a difference between a value of y k+1 of the controlled variable at a sampling time point of k+1 and a set-point value, y d , within a predetermined range, when ψ=1; and said adjusting will cause an absolute difference between a value of y k+1 of the controlled variable at a sampling time point of k+1 and a set-point value, y d , smaller an absolute difference between the y k and y d , when ψ=0.
2. The method according to claim 1 , wherein said SAC is a PI controller, PID controller or neural adaptive controller.
3. The method according to claim 1 , wherein said MPC is an artificial neural network model controller.