Method for setting model threshold of facility monitoring system
An exemplary embodiment of the present disclosure discloses a method of setting a model threshold value for detecting an anomaly of a facility monitoring system, the method including: acquiring sensor data output from each sensor; extracting a feature value for the sensor data of each sensor; acquiring output data by inputting input data including the extracted feature value to a trained neural network model; and comparing the input data and the output data and setting a threshold value for detecting an anomaly based on a calculated comparison result value.
1. A method of setting a model threshold value for detecting an anomaly of a facility monitoring system, the method comprising:
acquiring sensor data output from each sensor;
extracting two or more feature values for each of variables included in the sensor data of each sensor, wherein the two or more feature values for each of the variables comprises an upper control limit value and a lower control limit value;
acquiring output data by inputting input data to a trained neural network model to generate the output data, wherein the input data consists of a combination of the extracted feature value; and
comparing the input data and the output data to generate a calculated comparison result value, and determining a threshold value which is for detecting an anomaly, based on the calculated comparison result value;
updating the threshold value based on (i) a past threshold value of the trained neural network model corresponding to a past failure class, (ii) a past median value of device state indexes corresponding to a normal section, (iii) a current threshold value of the trained neural network model corresponding to a current failure class, (iv) a current median value of device state indexes corresponding to a normal section, and (v) a correction value for failure cost, when the current failure class corresponds to the past failure class.
2. The method of claim 1 , wherein the threshold value is an initial threshold value for determining the anomaly.
3. The method of claim 2 , wherein the initial threshold value is determined based on the calculated comparison result value of the input data and the output data without using past failure data.
4. The method of claim 3 , wherein the calculated comparison result value includes a Health Index (HI) indicating a device state index, and wherein the initial threshold value comprises a largest value among the HIs or one of a maximum value, a mean, and a minimum value.
5. The method of claim 2 , wherein the initial threshold value is determined based on the calculated comparison result value of the input data and the output data without a user input.
6. The method of claim 1 , wherein updating the threshold value is differently performed between when the number of the past median value is one and when the number of the past median value is two or more.
7. The method of claim 1 , wherein the extracting comprises extracting two or more feature values for each of variables included in sensor data which falls in a normal range within the sensor data of each sensor.
8. The method of claim 1 , wherein the extracting comprises extracting two or more feature values of an upper limit value, a lower limit value, a mean, standard deviation, covariance for the sensor data output from each sensor.
9. A computer program stored in a computer readable storage medium, wherein when the computer program is executed by one or more processors, the computer program performs following operations for setting a model threshold value, the operations comprising:
acquiring sensor data output from each sensor;
extracting two or more feature values for each of variables included in the sensor data of each sensor, wherein the two or more feature values for each of the variables comprises an upper control limit value and a lower control limit value;
acquiring output data by inputting input data to a trained neural network model to generate the output data, wherein the input data consists of a combination of the extracted feature values; and
comparing the input data and the output data to generate a calculated comparison result value, and determining a threshold value which is for detecting an anomaly, based on the calculated comparison result value;
updating the threshold value based (i) on a past threshold value of the trained neural network model corresponding to a past failure class, (ii) a past median value of device state indexes corresponding to a normal section, (iii) a current threshold value of the trained neural network model corresponding to a current failure class, (iv) a current median value of device state indexes corresponding to a normal section, and (v) a correction value for failure cost, when the current failure class corresponds to the past failure class.
10. A computing device for providing a method of setting a model threshold value, the computing device comprising:
a processor including one or more cores; and
a memory, wherein the processor:
acquires sensor data output from each sensor;
extracts two or more feature values for each of variables included in the sensor data of each sensor, wherein the two or more feature values for each of the variables comprises an upper control limit value and a lower control limit value;
acquires output data by inputting input data to a trained neural network model to generate the output data, wherein the input data consists of a combination of the extracted feature values; and
compares the input data and the output data to generate a calculated comparison result value, and determining a threshold value which is for detecting an anomaly, based on the calculated comparison result value;
wherein the threshold value is updated based on (i) a past threshold value of the trained neural network model corresponding to a past failure class, (ii) a past median value of device state indexes corresponding to a normal section, (iii) a current threshold value of the trained neural network model corresponding to a current failure class, (iv) a current median value of device state indexes corresponding to a normal section, and (v) a correction value for failure cost, when the current failure class corresponds to the past failure class.