Validation of operating plans and schedules using machine learning

1. A computer implemented method of industrial process control, comprising:

receiving in computer memory a dataset including initial process parameters representing operational data of a subject industrial process;

identifying a subset of the process parameters indicative of control data for controlling the subject industrial process, said identifying including:

(i) filtering the initial process parameters as a function of relative importance among the initial process parameters in describing variability of the operational data, wherein the filtering as a function of relative importance among the initial process parameters in describing the variability includes filtering the initial process parameters based on both: (i) contribution of each initial process parameter to a behavior of the operational data and (ii) contribution of each initial process parameter to one or more principal components of the operational data, wherein filtering the initial process parameters based on contribution of each initial process parameter to the one or more principal components comprises projecting the initial process parameters onto a space of the one or more principal components, constructing a hypervolume around an origin of the space, and removing process parameters in the hypervolume, the filtering resulting in a filtered dataset having process parameters above a threshold of importance to the behavior of the operational data;

(ii) grouping into one or more clusters the process parameters of the filtered dataset based on closeness of linear correlation among the process parameters of the filtered dataset, wherein each cluster includes grouped process parameters with close linear correlation; and

(iii) for each cluster resulting from the grouping:

determining a number of process parameters to extract from the cluster; and

individually extracting process parameters, representative of the cluster, from the grouped process parameters of the cluster, based on the determined number of process parameters and any one or combination of (a) a priority rating, (b) a measure of collinearity between the grouped process parameters within each given cluster of the determined clusters, (c) a measure of importance associated with the grouped process parameters, and (d) randomly or pseudo-randomly, wherein the extracting stores a respective individual indication of each individual process parameter extracted individually from each cluster and the process parameters extracted are the identified subset of the process parameters; and

automatically applying the identified subset of the process parameters controlling the subject industrial process.

2. The method of claim 1 , wherein the industrial process includes at least one of an optimized operational planning process, a scheduling process, a simulated chemical plant process, and an actual chemical plant process.

3. The method of claim 1 , wherein the operational data includes at least one of operating plan data and scheduling data.

4. The method of claim 1 , wherein the initial process parameters include at least one of a primal variable and a dual variable.

5. The method of claim 1 , wherein the initial process parameters include at least one of a process variable and a process constraint.

6. The method of claim 1 , further comprising constructing a principal component analysis (PCA) model that reduces dimensionality of the operational data of the dataset.

7. The method of claim 1 , wherein determining the number of process parameters to extract is based proportionally on at least one of (a) relative hyper-volumes of the clusters, (b) relative numbers of the grouped process parameters of the clusters, and (c) a uniform distribution between a plurality of the clusters.

8. The method of claim 1 , wherein at least one of:

the priority rating is based on information indicative of subject matter expertise, the information comprising at least one of (a) metadata and (b) embedded information;

the priority rating is determined based on domain knowledge;

the measure of collinearity is a measure of linear correlation of the grouped process parameters within a projected space; and

the measure of linear correlation of the grouped process parameters within the projected space increases as proximity between the grouped process parameters increases, and the measure of collinearity decreases as the grouped process parameters approach one or more edges of a cluster of the one or more clusters.

9. A computer system for industrial process control, the system comprising:

a processor operatively coupled to a data storage system, the processor configured to:

receive in computer memory a dataset including initial process parameters representing operational data of a subject industrial process;

identify a subset of the process parameters indicative of control data for controlling the subject industrial process, said identifying including:

(i) filtering the initial process parameters as a function of relative importance among the initial process parameters in describing variability of the operational data, wherein the filtering as a function of relative importance among the initial process parameters in describing the variability includes filtering the initial process parameters based on both: (i) contribution of each initial process parameter to a behavior of the operational data and (ii) contribution of each initial process parameter to one or more principal components of the operational data, wherein filtering the initial process parameters based on contribution of each initial process parameter to the one or more principal components comprises projecting the initial process parameters onto a space of the one or more principal components, constructing a hypervolume around an origin of the space, and removing process parameters in the hypervolume, the filtering resulting in a filtered dataset having process parameters above a threshold of importance to the behavior of the operational data;

(ii) grouping into one or more clusters the process parameters of the filtered dataset based on closeness of linear correlation among the process parameters of the filtered dataset, wherein each cluster includes grouped process parameters with close linear correlation; and

(iii) for each cluster resulting from the grouping:

determining a number of process parameters to extract from the cluster; and

individually extracting process parameters, representative of the cluster, from the grouped process parameters of the cluster, based on the determined number of process parameters and any one or combination of (a) a priority rating, (b) a measure of collinearity between the grouped process parameters within each given cluster of the determined clusters, (c) a measure of importance associated with the grouped process parameters, and (d) randomly or pseudo-randomly, wherein the extracting stores a respective individual indication of each individual process parameter extracted individually from each cluster and the process parameters extracted are the identified subset of the process parameters; and

automatically apply the identified subset of the process parameters controlling the subject industrial process.

10. The computer system of claim 9 , wherein the industrial process includes at least one of an optimized operational planning process, a scheduling process, a simulated chemical plant process, and an actual chemical plant process.

11. The computer system of claim 9 , wherein the operational data includes at least one of operating plan data and scheduling data.

12. The computer system of claim 9 , wherein the initial process parameters include at least one of a primal variable and a dual variable.

13. The computer system of claim 9 , wherein the initial process parameters include at least one of a process variable and a process constraint.

14. The computer system of claim 9 , wherein the processor is further configured to construct a principal component analysis (PCA) model that reduces dimensionality of the operational data of the dataset.

15. The computer system of claim 9 , wherein the processor is further configured to determine the number of process parameters to extract based proportionally on at least one of (a) relative hyper-volumes of the clusters, (b) relative numbers of the grouped process parameters of the clusters, and (c) a uniform distribution between a plurality of the clusters.

16. The computer system of claim 9 , wherein at least one of:

the priority rating is based on information indicative of subject matter expertise, the information comprising at least one of (a) metadata and (b) embedded information;

the priority rating is determined based on domain knowledge;

the measure of collinearity is a measure of linear correlation of the grouped process parameters within a projected space; and

the measure of linear correlation of the grouped process parameters within the projected space increases as proximity between the grouped process parameters increases, and the measure of collinearity decreases as the grouped process parameters approach one or more edges of a cluster of the one or more clusters.

17. A non-transitory computer-readable data storage medium comprising instructions causing a computer to:

receive in computer memory a dataset including initial process parameters representing operational data of a subject industrial process;

identify a subset of the process parameters indicative of control data for controlling the subject industrial process, said identifying including:

(i) filtering the initial process parameters as a function of relative importance among the initial process parameters in describing variability of the operational data, wherein the filtering as a function of relative importance among the initial process parameters in describing the variability includes filtering the initial process parameters based on both: (i) contribution of each initial process parameter to behavior of the operational data and (ii) contribution of each initial process parameter to one or more principal components of the operational data, wherein filtering the initial process parameters based on contribution of each initial process parameter to the one or more principal components comprises projecting the initial process parameters onto a space of the one or more principal components, constructing a hypervolume around an origin of the space, and removing process parameters in the hypervolume, the filtering resulting in a filtered dataset having process parameters above a threshold of importance to the behavior of the operational data;

(ii) grouping into one or more clusters the process parameters of the filtered dataset based on closeness of linear correlation among the process parameters of the filtered dataset, wherein each cluster includes grouped process parameters with close linear correlation; and

(iii) for each cluster resulting from the grouping:

determining a number of process parameters to extract from the cluster; and

individually extracting process parameters, representative of the cluster, from the grouped process parameters of the cluster, based on the determined number process parameters and any one or combination of (a) a priority rating, (b) a measure of collinearity between the grouped process parameters within each given cluster of the determined clusters, (c) a measure of importance associated with the grouped process parameters, and (d) randomly or pseudo-randomly, wherein the extracting stores a respective individual indication of each individual process parameter extracted individually from each cluster and the process parameters extracted are the identified subset of the process parameters; and

automatically apply the identified subset of the process parameters controlling the subject industrial process.

18. The non-transitory computer-readable data storage medium of claim 17 wherein the instructions cause the computer to:

(i) construct a principal component analysis (PCA) model that reduces dimensionality of the operational data of the dataset; and

(ii) determine the number of process parameters to extract based proportionally on at least one of (a) relative hyper-volumes of the clusters, (b) relative numbers of the grouped process parameters of the clusters, and (c) a uniform distribution between a plurality of the clusters.