Systems and methods for creating and using combined predictive models to control HVAC equipment

1. A heating, ventilation, or air conditioning (HVAC) system for a building, the HVAC system comprising:

HVAC equipment operable to affect an environmental condition in the building, a performance of the HVAC equipment characterized by a performance variable;

a controller configured to determine an operating point for the HVAC equipment and to operate the HVAC equipment at the operating point; and

a predictive modeling system comprising:

a first HVAC component model configured to generate a first component model prediction comprising a first predicted value of the performance variable for a particular future time;

a second HVAC component model configured to generate a second component model prediction comprising a second predicted value of the performance variable for the particular future time; and

one or more prediction combiners configured to combine the first and second component model predictions to form a combined model prediction, wherein the combined model prediction comprises another predicted value of the performance variable at the particular future time;

wherein the controller is configured to use the combined model prediction to optimize the operating point and to operate the HVAC equipment at the optimized operating point; and

one or more sensors configured to measure one or more measured variables associated with the HVAC equipment;

wherein the first and second HVAC component models are configured to generate the component model predictions as functions of the measured variables.

2. The HVAC system of claim 1 , wherein the operating point is a setpoint for the HVAC equipment and the performance variable indicates a predicted power consumption of the HVAC equipment at the setpoint.

3. The HVAC system of claim 1 , wherein the first and second HVAC component models have different functional forms and use different mathematical relationships to generate the first and second component model predictions.

4. The HVAC system of claim 1 , wherein the prediction combiners comprise an equal weighting combiner configured to generate the combined model prediction by calculating an average of the component model predictions.

5. The HVAC system of claim 1 , wherein the prediction combiners comprise a variance weighting combiner configured to:

identify a variance associated with each of the component model predictions;

assign a weight to each of the component model predictions based on the variance associated therewith; and

generate the combined model prediction by calculating a weighted average of the component model predictions using the assigned weights.

6. The HVAC system of claim 1 , wherein the prediction combiners comprise a trimmed mean combiner configured to:

create an initial set of the component model predictions;

identify one or more highest values of the component model predictions and one or more lowest values of the component model predictions;

create a filtered subset of the component model predictions by removing the identified component model predictions from the initial set; and

generate the combined model prediction by calculating an average of the component model predictions in the filtered subset.

7. The HVAC system of claim 6 , wherein the trimmed mean combiner is configured to use a variance weighting technique to calculate the average of the component model predictions in the filtered subset, the variance weighting technique comprising:

identifying a variance associated with each of the component model predictions in the filtered subset;

assigning a weight to each of the component model predictions based on the variance associated therewith; and

generating the combined model prediction by calculating a weighted average of the component model predictions in the filtered subset using the assigned weights.

8. The HVAC system of claim 1 , further comprising:

a model evaluator configured to calculate a score for each of the prediction combiners by comparing the combined model predictions generated by each of the prediction combiners to actual values of the performance variable; and

a combiner selector configured use the calculated scores to select one of the prediction combiners;

wherein the predictive modeling system is configured to use the selected prediction combiner to generate the combined model prediction.