Systems and methods for retraining outlier detection limits in a building management system

• US 10,120,375 B2
• Filed: 04/23/2015
• Issued: 11/06/2018
• Est. Priority Date: 04/23/2015
• Status: Active Grant

• Current Assignee: Tyco Fire & Security GmbH (Johnson Controls International plc)
• Original Assignee: Johnson Controls Technology Company (Johnson Controls International plc)
• Inventors: Przybylski, Andrew J., Wenzel, Michael J., Boettcher, Andrew J.
• Primary Examiner(s): Yi, Roy Y
• Assistant Examiner(s): Aiello, Jeffrey
• Application Number:
  US14/694,660
• Publication Number:
  US 20160313023A1
• Time in Patent Office: 1,293 Days
• Field of Search: 700275, 700276, 700291, 702179, 702182, 702183, 236 1
• CPC Class Codes: F24F 11/30 for purposes related to the operation of the system, e.g. for safety or monitoring
• CPC Class Codes: F24F 11/63 Electronic processing
• CPC Class Codes: F24F 2110/20 Humidity
• CPC Class Codes: G05B 2219/2614 HVAC, heating, ventillation, climate control
• CPC Class Codes: G05B 2219/32201 Build statistical model of past normal proces, compare with actual process
• CPC Class Codes: G05B 23/0297 Reconfiguration of monitoring system, e.g. use of virtual sensors; change monitoring method as a response to monitoring results

Abstract

A building management system (BMS) includes a controller that monitors performance values for a controlled process during a first time period relative to initial outlier detection limits and generates new outlier detection limits for the controlled process in response to a detected change in the controlled process during the first time period. The controller monitors the performance values relative to the new outlier detection limits during a second time period to detect outliers during the second time period. The controller calculates a confidence difference for an estimated confidence parameter based on a number of outliers detected using the new outlier detection limits during the second time period. The controller adjusts the new outlier detection limits in response to the confidence difference dropping below a threshold value.

Claims

1. A building management system comprising:

one or more sensors that measure a variable state or condition representing an output of a controlled process in the building management system;

building equipment that operate to affect the variable state or condition of a building environment measured by the one or more sensors; and

a controller in communication the building equipment and the one or more sensors, wherein the controller;

provides a control signal to the building equipment, the control signal causing the building equipment to operate to affect the variable state or condition of the building environment measured by the one or more sensors;

monitors performance values for the controlled process during a first time period relative to initial outlier detection limits;

generates new outlier detection limits for the controlled process in response to a detected change in the controlled process during the first time period;

outputs a user alert message comprising a fault event notification in response to the detected change in the controlled process;

monitors the performance values relative to the new outlier detection limits during a second time period to detect outliers during the second time period;

calculates a confidence difference for an estimated confidence parameter based on a number of outliers detected using the new outlier detection limits during the second time period, wherein the estimated confidence parameter is a ratio of a number of outliers detected during the second time period to a number of the performance values observed during the second time period; and

adjusts the new outlier detection limits in response to the confidence difference dropping below a threshold value.

2. The building management system of claim 1, wherein calculating the confidence difference comprises:

generating an estimated confidence parameter for each of the performance values observed during the second time period;

identifying a maximum and a minimum of the generated confidence parameters; and

calculating the confidence difference by subtracting the minimum confidence parameter from the maximum confidence parameter.

3. The building management system of claim 1, wherein in response to detecting the change in the controlled process, the controller resets the confidence difference and discards any performance values used to calculate the confidence difference prior to the detected change.

4. The building management system of claim 1, wherein adjusting the new outlier detection limits comprises:

identifying a reference time at which the confidence difference drops below a first threshold value;

storing the confidence difference at the reference time as a reference confidence difference; and

adjusting the new outlier detection limits in response to the confidence difference dropping below a second threshold value, wherein the second threshold value is a function of the reference confidence difference.

5. The building management system of claim 1, wherein adjusting the new outlier detection limits comprises:

identifying a reference time at which the confidence difference drops below a first threshold value;

storing the outlier detection limits at the reference time as reference outlier detection limits; and

maintaining the new outlier detection within a widened set of outlier detection limits based on the reference outlier detection limits.

6. The building management system of claim 1, wherein the performance values are process errors based on a difference between the output of the controlled process and a setpoint for the controlled process.

7. The building management system of claim 6, wherein the performance values are exponentially-weighted moving averages of the process errors.

8. The building management system of claim 1, wherein monitoring the performance values during the first time period comprises:

detecting outliers during the first time period using the initial outlier detection limits; and

detecting the change in the controlled process in response to a number of outliers detected during the first time period exceeding an outlier threshold.

9. The building management system of claim 8, wherein the controller calculates the outlier threshold as a function of the confidence difference.

10. The building management system of claim 1, wherein monitoring the performance values during the first time period comprises:

determining a current variance of the performance values during the first time period;

determining a previous variance of the performance values during a previous time period prior to the first time period; and

using the current variance of the performance values and the previous variance of the performance values to detect the change in the controlled process.

11. The building management system of claim 10, wherein detecting the change in the controlled process comprises:

using the current variance of the performance values and the previous variance of the performance values to generate a test statistic;

comparing the test statistic with a critical value; and

detecting the change in the controlled process in response to the critical value exceeding the test statistic.

12. A controller in a building management system, the controller comprising:

a communications interface that receives performance values indicating the performance of a controlled process in the building management system, and provides control signals to building equipment, the control signals causing the building equipment to operate to affect the controlled process;

an outlier detector that monitors performance values for the controlled process during a first time period relative to initial outlier detection limits and outputs a user alert message comprising a fault event notification in response to a detected change in the controlled process;

a control limit trainer that generates new outlier detection limits in response to the detected change in the controlled process during the first time period, wherein the outlier detector monitors the performance values relative to the new outlier detection limits during a second time period to detect outliers during the second time period;

a confidence difference calculator that calculates a confidence difference for an estimated confidence parameter based a number of outliers detected using the new outlier detection limits during the second time period; and

a stationary process detector that detects when the confidence difference drops below a threshold value during the second time period, wherein the control limit trainer adjusts the new outlier detection limits in response to the confidence difference dropping below the threshold value;

wherein the confidence difference calculator generates an estimated confidence parameter for each of the performance values observed during the second time period, and the confidence parameter is a ratio of a number of outliers detected during the second time period to a number of the performance values observed during the second time period.

13. The controller of claim 12, wherein the confidence difference calculator:

identifies a maximum and a minimum of the generated confidence parameters; and

calculates the confidence difference by subtracting the minimum confidence parameter from the maximum confidence parameter.

14. The controller of claim 12, wherein the outlier detector detects outliers during the first time period using the initial outlier detection limits;

the controller further comprising a process change detector that detects the change in the controlled process in response to a number of outliers detected during the first time period exceeding an outlier threshold.

15. The controller of claim 12, further comprising a lower variance detector that determines a current variance of the performance values during the first time period and determines a previous variance of the performance values during a previous time period prior to the first time period;

the controller further comprising a process change detector that uses the current variance of the performance values and the previous variance of the performance values to detect the change in the controlled process.

16. The controller of claim 15, wherein the lower variance detector:

uses the current variance of the performance values and the previous variance of the performance values to generate a test statistic;

compares the test statistic with a critical value; and

identifies a lower variance event in response to the critical value exceeding the test statistic;

wherein the process change detector detects the change in the controlled process in response to the lower variance event.

17. The controller of claim 12, wherein the performance values are exponentially-weighted moving averages of a process error, wherein the process error is a difference between an output of the controlled process and a setpoint for the controlled process.

18. The controller of claim 12, wherein in response to detecting the change in the controlled process, the confidence difference calculator resets the confidence difference and discards any performance values used to calculate the confidence difference prior to the detected change.

19. The controller of claim 12, wherein adjusting the new outlier detection limits comprises:

identifying a reference time at which the confidence difference drops below the threshold value;

storing the outlier detection limits at the reference time as reference outlier detection limits; and

maintaining the new outlier detection within a widened set of outlier detection limits based on the reference outlier detection limits.