Building thermal control techniques

1. A method comprising the steps of:

constructing an algebra and differential equations model of heat transfer in a building based on available training data and physical system characteristics;

carrying out a hybrid calibration process to iteratively calibrate both time-insensitive and time-sensitive parameters of said algebra and differential equations model so as to obtain parameter vectors;

applying vector auto-regression to said parameter vectors to predict values of said parameters for a future time period;

configuring a model predictive controller of a building heating, ventilating, and air conditioning system according to a predictive model, based on said predicted values of said parameters for said future time period; and

controlling said building heating, ventilating, and air conditioning system in accordance with said predictive model,

wherein said step of carrying out said hybrid calibration process comprises:

for each sub time period of a predetermined time period of said available training data, training said time-insensitive and time-sensitive parameters together;

based on said training of said time-insensitive and time-sensitive parameters together, training said time-insensitive parameters with said predetermined time period of said available training data, with said time sensitive parameters fixed, to obtain updated time-insensitive parameters; and

using said updated time-insensitive parameters to train an updated set of time-sensitive parameters for each of said sub time periods,

wherein said step of training said time-insensitive and time-sensitive parameters together for each sub time period of said predetermined time period of said available training data comprises:

making an initial estimate of said time-insensitive and time-sensitive parameters;

using a genetic algorithm to determine a parameter fitting of said time-insensitive and time-sensitive parameters for said each sub time period of said predetermined time period of said available training data; and

repeating said steps of making said initial estimate of said time-insensitive and time-sensitive parameters and using said genetic algorithm for said each sub time period until convergence is achieved;

said step of training said time-insensitive parameters to obtain updated time-insensitive parameters comprises:

making an initial estimate of said time insensitive parameters;

using said genetic algorithm to determine a parameter fitting of said time-insensitive parameters for said predetermined time period of said available training data; and

repeating said steps of making said initial estimate of said time insensitive parameters and using said genetic algorithm to determine said parameter fitting of said time-insensitive parameters until convergence is achieved; and

said step of using said updated time-insensitive parameters to train said updated set of time-sensitive parameters for each of said sub time periods comprises:

making an initial estimate of said time sensitive parameters;

using said genetic algorithm to determine a parameter fitting for said time sensitive parameters for said each sub time period of said predetermined time period of said available training data; and

repeating said steps of making said initial estimate of said time sensitive parameters and using said genetic algorithm for said time sensitive parameters until convergence is achieved.

2. The method of claim 1 , wherein:

said step of carrying out said hybrid calibration process is implemented by a hybrid calibration sub-module of a modeling module, embodied on a non-transitory computer-readable medium, executing on at least one hardware processor; and

said step of applying said vector auto-regression to said parameter vectors is implemented by a vector auto-regression sub-module of said modeling module, embodied on said non-transitory computer-readable medium, executing on said at least one hardware processor.