Building management system with self-optimizing control modeling framework

1. A building management system comprising:

a plant comprising building equipment operable to affect a variable state or condition of a building, the plant operating at a cost indicated by a performance variable; and

a self-optimizing controller configured to:

provide a manipulated variable as a control input to the plant and receive an output variable as a feedback from the plant;

generate a performance variable model that defines the performance variable as a function of the output variable and the manipulated variable;

generate an output variable model that defines the output variable as a function of the manipulated variable;

use the performance variable model to determine a gradient of the performance variable with respect to at least one of the output variable and the manipulated variable;

use the output variable model to determine a gradient of the output variable with respect to the manipulated variable;

generate a model of a self-optimizing variable based on the gradient of the performance variable model and the gradient of the output variable model; and

operate the building equipment of the plant to affect the variable state or condition of the building based on a value of the self-optimizing variable defined by the self-optimizing variable model.

2. The system of claim 1 , wherein the performance variable model and the output variable model are generated using a regression technique.

3. The system of claim 2 , wherein the regression technique is any of a linear regression, a Taylor expansion, a support vector machine, a K-nearest neighbor regression, a partial least square fit regression, a regression tree, a generalized linear model, a neural network, and a random forest.

4. The system of claim 1 , wherein determining the gradient of the performance variable with respect to at least one of the output variable and the manipulated variable comprises:

determining a partial derivative of the performance variable with respect to the manipulated variable; and

determining another partial derivative of the performance variable with respect to the output variable.

5. The system of claim 1 , wherein the self-optimizing controller is configured to perturb the manipulated variable with a step input to determine values of the manipulated variable over a time duration and provide the values of the manipulated variable over the time duration to the plant as the control input.

6. The system of claim 1 , wherein the performance variable model is generated based on a set of values of the performance variable, the manipulated variable, and the output variable.

7. The system of claim 1 , wherein the output variable model is generated based on a set of values of the output variable and the manipulated variable.

8. The system of claim 1 , wherein the self-optimizing controller is configured to generate control signals for the building equipment such that the self-optimizing variable is driven toward zero.

9. The system of claim 1 , wherein the output variable of the plant is influenced by a disturbance, and wherein the performance variable is influenced by the disturbance.

10. A self-optimizing controller for building equipment of a plant, the controller configured to:

provide a manipulated variable as a control input to the plant and receive an output variable as a feedback from the plant;

generate a performance variable model that defines the performance variable as a function of the manipulated variable;

generate an output variable model that defines the output variable as a function of the manipulated variable;

use the performance variable model to determine a gradient of the performance variable with respect to at least one of the output variable and the manipulated variable;

use the output variable model to determine a gradient of the output variable with respect to the manipulated variable;

generate a self-optimizing variable based on the gradient of the performance variable model and the gradient of the output variable model; and

operate the building equipment of the plant to affect a variable state or condition of the building based on the value of the self-optimizing variable defined by the self-optimizing variable model.

11. The controller of claim 10 , wherein the performance variable model and the output variable model are generated using a regression technique.

12. The controller of claim 11 , wherein the regression technique is any of a linear regression, a Taylor expansion, a support vector machine, a K-nearest neighbor regression, a partial least square fit regression, a regression tree, a generalized linear model, a neural network, and a random forest.

13. The controller of claim 10 , wherein determining the gradient of the performance with respect to at least one of the output variable and the manipulated variable comprises:

determining a partial derivative of the performance variable with respect to the manipulated variable; and

determining another partial derivative of the performance variable with respect to the output variable.

14. The controller of claim 10 , wherein the controller is configured to perturb the manipulated variable with a step input to determine values of the manipulated variable over a time duration and provide the values of the manipulated variable over the time duration to the plant as the control input.

15. The controller of claim 10 , wherein the performance variable model is generated based on a set of values of the performance variable, the manipulated variable, and the output variable.

16. The controller of claim 10 , wherein the output variable model is generated based on a set of values of the output variable and the manipulated variable.

17. The controller of claim 10 , wherein the controller is further configured to generate control signals for the building equipment such that the self-optimizing variable is driven toward zero.

18. A method for performing self-optimizing control on a plant, the method comprising:

providing a manipulated variable as a control input to the plant and receiving an output variable as a feedback from the plant;

generating a performance variable model that defines the performance variable as a function of the output variable and the manipulated variable;

generating an output variable model that defines the output variable as a function of the manipulated variable;

using the performance variable model to determine a gradient of the performance variable with respect to at least one of the output variable and the manipulated variable;

using the output variable model to determine a gradient of the output variable with respect to the manipulated variable;

generating a model of a self-optimizing variable based on the gradient of the performance variable model and the gradient of the output variable model; and

operating the building equipment of the plant to affect the variable state or condition of the building based on a value of the self-optimizing variable defined by the self-optimizing variable model.

19. The method of claim 18 , wherein using the performance variable model to determine a gradient of the performance variable comprises:

determining a partial derivative of the performance variable with respect to the manipulated variable; and

determining another partial derivative of the performance variable with respect to the output variable.

20. The method of claim 18 , further comprising generating control signals for the building equipment such that the self-optimizing variable is driven toward zero.