Building management system with efficient model generation for system identification

1. A building management system comprising:

building equipment operable to heat or cool a building and generate training data relating to behavior of a building system; and

a controller configured to perform a system identification process to identify one or more parameters of a system model that predicts the behavior of the building system, the one or more parameters comprising at least one of one or more model parameters or one or more Kalman gain parameters, the system identification process comprising:

generating a prediction error function based on the training data and the system model;

generating a plurality of initial guesses of the one or more parameters;

for each initial guess, running an optimization problem of the prediction error function for a first group of iterations;

after the first group of iterations, discarding a portion of the initial guesses based on one or more criteria and ranking a remaining portion of the initial guesses;

running the optimization problem of the prediction error function for a top-ranked initial guess of the remaining portion to local optimality or for a second group of iterations to identify a first set of values of the one or more parameters;

determining whether the first set of values of the one or more parameters satisfies a condition; and

in response to a determination that the first set of values satisfies the condition, identifying the one or more parameters as having the first set of values;

wherein the controller is configured to operate the building equipment to heat or cool the building by performing a predictive control process that uses the system model with the one or more parameters identified by the system identification process.

2. The building management system of claim 1 , wherein the controller is further configured to, in response to determining that the first set of values does not satisfy the condition:

discard the top-ranked initial guess;

run the optimization problem for a new top-ranked initial guess to local optimality or for the second group of iterations to identify a second set of values of the one or more parameters;

determine whether the second set of values satisfies the condition; and

in response to a determination that the second set of values satisfies the condition, select the second set of values for the one or more parameters.

3. The building management system of claim 1 , wherein the controller is configured to discard initial guesses that violate physical laws relating to the one or more parameters.

4. The building management system of claim 3 , wherein determining that the first set of values satisfies the condition comprises determining that the first set of values conforms with the physical laws relating to the one or more parameters.

5. The building management system of claim 1 , wherein the controller is configured to discard the portion of the initial guesses based on one or more criteria by determining that a first initial guess and a second initial guess lead toward a same local optimum after the first group of iterations and, in response, discarding the first initial guess.

6. The building management system of claim 1 , wherein the controller is configured to discard the portion of the initial guesses based on one or more criteria by discarding initial guesses that lead to an unstable, uncontrollable, or unobservable system after the first group of iterations.

7. The building management system of claim 1 , wherein the controller is configured to discard the portion of the initial guesses based on one or more criteria by discarding initial guesses that lead to a matrix with a condition number greater than a threshold number.

8. The building management system of claim 1 , wherein each of the plurality of initial guesses provides a stable system.

9. The building management system of claim 1 , wherein the controller is configured to detect and remove saturation data from the training data.

10. The building management system of claim 1 , wherein the prediction error function comprises a multi-step ahead prediction error function.

11. A method for generating a system model for a building system and controlling building equipment in accordance with the system model, comprising:

operating the building equipment to generate training data relating to behavior of the building system;

defining a prediction error function based on the training data and the system model;

identifying values for one or more parameters of the system model by optimizing the prediction error function, wherein optimizing the prediction error function comprises:

generating a plurality of initial guesses of the one or more parameters;

for each initial guess, running an optimization problem of the prediction error function for a first group of iterations;

after the first group of iterations, discarding a portion of the initial guesses based on one or more criteria and ranking a remaining portion of the initial guesses;

running the optimization problem of the prediction error function for a top-ranked initial guess of the remaining portion to local optimality or for a second group of iterations to identify a first set of values of the one or more parameters;

determining whether the first set of values of the one or more parameters satisfies a condition; and

in response to a determination that the first set of values satisfies the condition, defining the one or more parameters as having the first set of values; and

operating the building equipment to heat or cool the building system by applying the system model in a predictive controller to generate control inputs for the building equipment.

12. The method of claim 11 , further comprising, in response to determining that the first set of values does not satisfy the condition:

discarding the top-ranked initial guess;

running the optimization problem of the prediction error function for a new top-ranked initial guess to local optimality or for the second group of iterations to identify a second set of values of the one or more parameters;

determining whether the second set of values satisfies the condition; and

in response to a determination that the second set of values satisfies the condition, selecting the second set of values for the one or more parameters.

13. The method of claim 11 , comprising discarding initial guesses that violate physical laws relating to the one or more parameters and the behavior of the building system.

14. The method of claim 13 , wherein determining that the first set of values satisfies the condition comprises determining that the first set of values conforms with the physical laws.

15. The method of claim 11 , wherein discarding the portion of the initial guesses based on one or more criteria comprises determining that a first initial guess and a second initial guess lead toward a same local optimum after the first group of iterations and, in response, discarding the first initial guess.

16. The method of claim 11 , wherein discarding the portion of the initial guesses based on one or more criteria comprises discarding initial guesses that lead to an unstable, uncontrollable, or unobservable system after the first group of iterations.

17. The method of claim 11 , wherein discarding the portion of the initial guesses based on one or more criteria comprises discarding initial guesses that lead to a matrix with a condition number greater than a threshold number after the first group of iterations.

18. The method of claim 11 , wherein each of the plurality of initial guesses provides a stable system.

19. The method of claim 11 , comprising detecting and removing saturation data from the training data.

20. A method for generating a system model for a building system and controlling building equipment in accordance with the system model, comprising:

operating the building equipment to generate training data relating to behavior of the building system;

defining a prediction error function based on the training data and the system model;

identifying values for one or more parameters of the system model based on the prediction error function by:

generating a plurality of initial guesses of the one or more parameters;

for each initial guess, performing a first group of iterations using the prediction error function;

after the first group of iterations, discarding a portion of the initial guesses based on one or more criteria and ranking a remaining portion of the initial guesses;

performing a second group of iterations using the prediction error function for a top-ranked initial guess of the remaining portion to identify a first set of values of the one or more parameters;

determining whether the first set of values of the one or more parameters satisfies a condition; and

in response to a determination that the first set of values satisfies the condition, defining the one or more parameters as having the first set of values; and

operating the building equipment to heat or cool the building system by applying the system model in a predictive controller to generate control inputs for the building equipment.