Estimating initial states of a system model for controlling an industrial process

A device for controlling an industrial process, based on an estimated initial state and a merged Mixed Logical Dynamic (MLD) system model of the industrial process, the merged MLD system model being a combination of MLD subsystem models representing sub-processes of the industrial process, the device comprising a processor configured to:

• optimize, as a function of state variables x, an objective function J comprising a difference between observed quantities y and outputs ŷ of the merged MLD system model for the industrial process, the optimization being performed over a number of time steps in the past and being subject to constraints defined by dynamics of the merged MLD system model;
• retain optimizing values x0 of the state variables at a time step as the estimated initial state, the time step being an optimized time step at which the optimizing values x0 of the state variables are estimated to occur;
• exclude the intermediary state variable from the optimization;
• and
• simulate a value of the intermediary state variable at the time step by evaluating the MLD system model, and retain the simulated value in the estimated initial state x0.

A non-transitory computer-readable recording medium having a program recorded thereon that causes a processor of a computer processing device to estimate initial states of a system model for an industrial process with a plurality of sub-processes, the program causing the processor to execute operations comprising:

• optimizing, as a function of state variables x, an objective function J including a difference between observed quantities y and outputs ŷ of a merged Mixed Logical Dynamic (MLD) system model for the industrial process, the optimization being performed over a number of time steps in the past and being subject to constraints defined by dynamics of the merged MLD system model, wherein the merged MLD system model is a combination of MLD subsystem models representing the sub-processes of the industrial process;
• retaining optimizing values x0 of the state variables at a time step as an estimated initial state for control, based on the merged MLD system model, of the industrial process, wherein the state variables include an intermediary state variable, and the time step is an optimized time step at which the optimizing values x0 of the state variables are estimated to occur;
• excluding the intermediary state variable from the optimization;
• and
• simulating a value of the intermediary state variable at the time step by evaluating the MLD system model, and retaining the simulated value in the estimated initial state x0.