Systems and methods for prediction model update scheduling for building equipment
A building system includes building equipment operable to consume one or more resources and a control system configured to generate, based on a prediction model, predictions of a load on the building equipment or a price of the one or more resources for a plurality of time steps in an optimization period, solve, based on the predictions, an optimization problem to generate control inputs for the equipment that minimize a predicted cost of consuming the resources over the optimization period, control the building equipment to operate in accordance with the control inputs, monitor an error metric that characterizes an error between the predictions and actual values of the at least one of the load on the building equipment or the price of the one or more resources during the optimization period, detect an occurrence of a trigger condition, and in response to detecting the trigger condition, update the prediction model.
1. A building system comprising:
building equipment operable to consume one or more resources and affect a variable state or condition of a building, the resources provided by one or more utility systems;
a control system configured to:
generate, based on a prediction model, predictions of at least one of a load on the building equipment or a price of the one or more resources for a plurality of time steps in an optimization period;
solve, based on the predictions, an optimization problem to generate control inputs for the equipment that minimize a predicted cost of consuming the one or more resources over the optimization period;
control the building equipment to operate in accordance with the control inputs;
monitor an error metric that characterizes an error between the predictions and actual values of the at least one of the load on the building equipment or the price of the one or more resources during the optimization period;
update a control region for the error metric in accordance with a change in a statistical property of the error metric;
detect an occurrence of a trigger condition by comparing the error metric to the control region; and
in response to detecting the occurrence of the trigger condition, update the prediction model to generate an updated prediction model;
generate updated control inputs for the building equipment based on the updated prediction model; and
control the building equipment to operate in accordance with the updated control inputs.
2. The building system of claim 1 , wherein the control system is configured to calculate the error metric as a coefficient of variation weighted mean absolute prediction error of the predictions.
3. The building system of claim 2 , wherein the control system is configured to calculate the coefficient of variation weighted mean absolute prediction error using a user-selectable weighting.
4. The building system of claim 1 , wherein the control system is configured to calculate the error metric as a smoothed current prediction error.
5. The building system of claim 1 , wherein the trigger condition occurs when the error metric is outside the control region.
6. The building system of claim 5 , wherein the trigger condition occurs when the error metric is outside the control region for at least a threshold duration.
7. The building system of claim 5 , wherein the control system is configured to determine at least one of an upper limit of the control region or a lower limit of the control region as a based on the statistical property of the error metric.
8. The building system of claim 7 , wherein the control system is configured to update the statistical property of the error metric in response to updating the prediction model.
9. A method for controlling a building system, the method comprising:
operating building equipment to consume one or more resources and affect a variable state or condition of a building, the resources provided by one or more utility systems;
generating, based on a prediction model, predictions of at least one of a load on the building equipment or a price of the one or more resources for a plurality of time steps in a time period;
generating, based on the predictions, control inputs for the equipment that manage a predicted cost of consuming the one or more resources over the time period;
controlling the building equipment to operate in accordance with the control inputs;
monitoring an error metric that characterizes an error between the predictions and actual values of the at least one of the load on the building equipment or the price of the one or more resources during the time period;
updating a control region for the error metric in accordance with a change in a statistical property of the error metric;
detecting an occurrence of a trigger condition by comparing the error metric to the control region;
in response to detecting the occurrence of the trigger condition, updating the prediction model to generate an updated prediction model;
generating updated control inputs for the building equipment based on the updated prediction model; and
controlling the building equipment to operate in accordance with the updated control inputs.
10. The method of claim 9 , comprising calculating the error metric as a coefficient of variation weighted mean absolute prediction error of the predictions.
11. The method of claim 10 , comprising calculating the coefficient of variation weighted mean absolute prediction error using a user-selectable weighting.
12. The method of claim 9 , comprising calculating the error metric as a smoothed current prediction error.
13. The method of claim 9 , wherein the trigger condition occurs when the error metric is outside the control region.
14. The method of claim 13 , wherein the trigger condition occurs when the error metric is outside the control region for at least a threshold duration.
15. The method of claim 13 , comprising determining at least one of an upper limit of the control region or a lower limit of the control region based on the statistical property of the error metric.
16. The method of claim 15 , comprising the updating the statistical property of the error metric in response to updating the prediction model.
17. A central plant comprising:
a plurality of subplants operable to consume, generate, or store one or more resources, at least one of the resources provided by a utility system at a utility rate;
a control system configured to:
generate, based on one or more prediction models, predictions of at least one of a load on a campus served by the central plant or the utility rate for a plurality of time steps in an optimization period;
solve, based on the predictions, an optimization problem to generate control inputs for the subplants that minimize a predict cost of consuming the resource provided by the utility system over the optimization period;
control the subplants to operate in accordance with the control inputs;
monitor an error metric that characterizes an error between the predictions and actual values of the at least one of the load on the campus served by the central plant or the utility rate;
detect an occurrence of a trigger condition, the trigger condition dynamically updated in accordance with a time-varying statistical property of the error metric;
in response to detecting the trigger condition, update the one or more prediction models;
control the plurality of subplants in accordance with an optimization strategy generated based on the one or more prediction models.
18. The central plant of claim 17 , wherein the control system is configured to calculate the error metric as a coefficient of variation weighted mean absolute prediction error of the predictions using a user-selectable weighting.
19. The central plant of claim 17 , wherein the control system is configured to calculate the error metric as a smoothed current prediction error.
20. The central plant of claim 17 , wherein the control system is configured to generate a graphical user interface comprising a visualization of the error metric and the trigger condition.