Site-wide optimization for mixed regression models and mixed control variables
A computer-implemented method for site-wide prediction optimization includes training a plurality of a mixed type of regression models with a mixed type of control variables for identifying control set-points of a site-wide operation. A decision tree regression model is trained to predict a status of the plurality of initial set-points for non-linear regression functions. The decision tree regression model is reformulated into a mixed-integer linear program (MILP) and solved by an MILP solver to find a global solution. An MILP surrogate is determined for a nonlinear optimization problem to provide a best solution for one or more of the non-linear regression functions using the best solution as a starting point for solving non-linear regression functions, and a set-point of the mixed control variables is recommended to control a throughput of the site-wide operation by executing a decomposition operation or a federated learning algorithm.
1 . A computer-implemented method for a site-wide prediction optimization of an industrial plant, the computer-implemented method comprising:
training a plurality of a mixed type of regression models with a mixed type of control variables, to operate a plurality of feed forward functions;
identifying control set-points of a site-wide industrial plant operation for the mixed type of control variables, based on the training of the plurality of the mixed type of regression models;
training a decision tree regression model to predict a status of the identified control set-points for the plurality of feed forward functions;
reformulating the decision tree regression model into a mixed-integer linear program (MILP) and solving the MILP with an MILP solver to find a global solution;
determining an MILP surrogate for a nonlinear optimization problem to provide an optimal solution for the plurality of feed forward functions;
solving, using the optimal solution as a starting point, the nonlinear optimization problem;
finding the global solution based on the solving of the nonlinear optimization problem, wherein the global solution corresponds to optimized control set-points of the identified control set-points; and
automatically performing, based on the optimized control set-points, a real-time adjustment of the mixed type of control variables to maximize a throughput of the site-wide industrial plant operation by executing a decomposition operation or a federated learning algorithm.
2 . The computer-implemented method of claim 1 , further comprising providing the optimal solution from the MILP surrogate as a starting point for performing a primal-dual operation.
3 . The computer-implemented method of claim 1 , further comprising recommending the optimized control set-points for the mixed type of control variables, wherein the recommending includes recommending one or more control set-points of the optimized control set-points for one or more of a continuous variable, an integer variable, or a categorical variable.
4 . The computer-implemented method of claim 1 , wherein the training of the plurality of the mixed type of regression models includes selecting two or more model types comprising piece-wise linear models, nonlinear models, or black-box models.
5 . The computer-implemented method of claim 4 , further comprising performing the decomposition operation by defining the two or more model types of the plurality of the mixed type of regression models as two subproblems for recommending a control set-point of the optimized control set-points, of each of the mixed type of control variables.
6 . The computer-implemented method of claim 5 , wherein
the performing of the decomposition operation comprises a temporal decomposition of a problem into the two subproblems, and
each subproblem of the two subproblems is related to either a regression function or a convex quadratic program.
7 . The computer-implemented method of claim 1 , wherein the training the decision tree regression model includes randomly generating samples of a controllable variable xi and estimating a function f (xi) of the controllable variable.
8 . The computer-implemented method of claim 1 , wherein
the site-wide industrial plant operation comprises an oil sands production facility including a network of multiple plants with each plant having a set of inputs and outputs, and
the computer-implemented method further comprises estimating an operation of each process of the network of the multiple plants using a regression function.
9 . The computer-implemented method of claim 8 , further comprising arranging each plant of the multiple plants of the network in a federated learning framework to preserve one or more of an objective information, a constraint information, or a private information.
10 . The computer-implemented method of claim 8 , further comprising:
initializing, by an automated central operator, a plurality of set-points for every process of the network of the multiple plants, a vector value, and an index;
sending, to each plant of the multiple plants, particular initialized set-points from the plurality of set-points, the vector value, and the index associated with the particular initialized set-points; and
receiving by the automated central operator an updated vector value after each plant of the multiple plants performs a self-optimization problem from each plant of the multiple plants.
11 . A computer-implemented method for site-wide prediction optimization of an industrial plant, the computer-implemented method comprising:
training a plurality of a mixed type of regression models with a mixed type of control variables, to operate a plurality of feed forward functions;
performing a decomposition operation of a site-wide optimization operation by defining two or more model types of the mixed type of regression models and the mixed type of control variables into respective suboperations;
executing, for each suboperation of the respective suboperations, a plurality of iterations of the mixed type of regression models until a predefined convergence is obtained;
performing, based on the executing of the plurality of iterations, an optimization of each suboperation of the respective suboperations, wherein the performing of the optimization includes:
identifying control set-points of a site-wide industrial plant operation for the mixed type of control variables based on the training of the plurality of the mixed type of regression models,
training a decision tree regression model to predict a status of the identified control set-points for the plurality of feed forward functions,
reformulating the decision tree regression model into a mixed-integer linear program (MILP) and solving the MILP with an MILP solver to find a global solution,
determining an MILP surrogate for a nonlinear optimization problem to provide an optimal solution for the plurality of feed forward functions,
solving, using the optimal solution as a starting point, the nonlinear optimization problem, and
finding the global solution based on the solving of the nonlinear optimization problem, wherein the global solution corresponds to optimized control set-points of the identified control set-points; and
automatically performing a real-time adjustment of the mixed type of control variables based on the optimization of each suboperation of the respective suboperations of the industrial plant.
12 . The computer-implemented method of claim 11 , wherein the performing of the optimization further comprises a non-convex optimization of each suboperation of the respective suboperations to designate the optimal solution having a lowest loss from among two or more sub-optimal solutions.
13 . The computer-implemented method of claim 12 , wherein the non-convex optimization includes selecting one or more of prediction models and operational constraints for each suboperation of the respective suboperations.
14 . The computer-implemented method of claim 11 , further comprising training the decision tree regression model to learn a nonlinear function to predict the status of the identified control set-points for the plurality of feed forward functions of the site-wide optimization operation.
15 . The computer-implemented method of claim 11 , wherein
performing the site-wide optimization operation is based on the training of the plurality of the mixed type of regression models including at least two of piece-wise linear models, nonlinear models, or black-box models, and the mixed type of control variables, and
the site-wide optimization operation is performed in parallel for at least two of the respective suboperations.
16 . The computer-implemented method of claim 15 , further comprising providing a secure platform with privacy-preserving functions to determine the optimized control set-points for a plurality of client nodes operating in a federated learning framework.
17 . A computing device configured to perform site-wide prediction optimization of an industrial plant, the computing device comprising:
a processor;
a memory coupled to the processor, the memory storing instructions to cause the processor to perform acts comprising:
training a plurality of a mixed type of regression models with a mixed type of control variables, to operate a plurality of feed forward functions;
identifying control set-points of a site-wide industrial plant operation for the mixed type of control variables, based on the training of the plurality of the mixed type of regression models;
training a decision tree regression model to predict a status of the identified control set-points for the plurality of feed forward functions;
reformulating the decision tree regression model into a mixed-integer linear program (MILP) and solving the MILP by an MILP solver to find a global solution;
determining an MILP surrogate for a nonlinear optimization problem to provide an optimal solution for the plurality of feed forward functions;
solving, using the optimal solution as a starting point, the nonlinear optimization problem;
finding the global solution based on the solving of the nonlinear optimization problem, wherein the global solution corresponds to optimized control set-points of the identified control set-points; and
automatically performing, based on the optimized control set-points, a real-time adjustment of the mixed type of control variables to maximize a throughput of the site-wide industrial plant operation by executing a decomposition operation or a federated learning algorithm.
18 . The computing device of claim 17 , wherein the instructions further cause the processor to perform an additional act of providing the optimal solution from the MILP surrogate as a starting point for performing a primal-dual operation.
19 . The computing device of claim 17 , wherein
the instructions further cause the processor to perform an additional act of recommending the optimized control set-points for the mixed type of control variables, and
the recommending includes recommending one or more control set-points of the optimized control set-points for one or more of a continuous variable, an integer variable, or a categorical variable.
20 . The computing device of claim 17 , wherein the training of the mixed type of regression models includes selecting two or more model types comprising piece-wise linear models, nonlinear models, or black-box models.
21 . The computing device of claim 17 , wherein the instructions further cause the processor to perform an additional act comprising performing the decomposition operation by defining two or more model types of the mixed type of regression models as two subproblems for recommending a control set-point of the optimized control set-points of each of the mixed type of control variables.
22 . The computing device of claim 17 , wherein the decomposition operation performed comprises a temporal decomposition of a problem into two subproblems.
23 . A non-transitory computer readable storage medium tangibly embodying a computer readable program code having computer readable instructions that, when executed, causes a computer device to carry out a method of site-wide prediction optimization of an industrial plant, the method comprising:
training a plurality of a mixed type of regression models, with mixed type of control variables, to operate a plurality of feed forward functions;
performing a decomposition operation of a site-wide optimization operation by defining two or more model types of the mixed type of regression models and the mixed type of control variables into respective suboperations;
executing, for each suboperation of the respective suboperations, a plurality of iterations of the mixed type of regression models until a predefined convergence is obtained;
performing, based on the executing of the plurality of iterations, an optimization of each suboperation of the respective suboperations, wherein the performing of the optimization includes:
identifying control set-points for the mixed type of control variables of a site-wide industrial plant operation based on the training of the plurality of the mixed type of regression models,
training a decision tree regression model to predict a status of the identified control set-points for the plurality of feed forward functions,
reformulating the decision tree regression model into a mixed-integer linear program (MILP) and solving the MILP with an MILP solver to find a global solution,
determining an MILP surrogate for a nonlinear optimization problem to provide an optimal solution for the plurality of feed forward functions,
solving, using the optimal solution as a starting point, the nonlinear optimization problem, and
finding the global solution based on the solving of the nonlinear optimization problem, wherein the global solution corresponds to optimized control set-points of the identified control set-points; and
automatically performing a real-time adjustment of the mixed type of control variables based on the optimization of each suboperation of the respective suboperations.
24 . The non-transitory computer readable storage medium according to claim 23 , wherein the method further comprises performing a non-convex optimization of each suboperation of the respective suboperations to designate the optimal solution having a lowest loss from among two or more sub-optimal solutions.
25 . The non-transitory computer readable storage medium according to claim 23 , wherein:
performing the site-wide optimization operation is based on the training of the plurality of mixed type of regression models including at least two of piece-wise linear models, nonlinear models, or black-box models, and the mixed type of control variables, and
the site-wide optimization operation is performed in parallel for at least two of the respective suboperations.