ANOMALY DETECTION IN MULTIPLE CORRELATED SENSORS
Embodiments include methods, systems and computer program products for detecting an anomaly in data provided by each one of a plurality of correlated sensors. Aspects include receiving time series data sequences from each one of a plurality of correlated sensors, determining a numeric representation for each one of the time series data sequences, determining an anomaly score for each one of the time series data sequences using the determined numeric representation for each one of the time series data sequences, and determining a distribution of the determined anomaly scores under normal conditions.
1 . A method for detecting an anomaly in data provided by each one of a plurality of correlated sensors, the method comprising:
receiving from each one of the plurality of correlated sensors a corresponding time series data sequence, each data sequence representing a plurality of data values sensed by a corresponding one of the plurality of correlated sensors at a sampling frequency, each of the data values of each data sequence being sensed at a particular point in time in the time series data sequence, wherein the correlated sensors have a correlation related to a common measured parameter or a relative proximity to one another;
determining a numeric representation for each one of the time series data sequences, wherein on a condition that the sampling frequency is not the same for each one of the plurality of correlated sensors, a vector of statistics for each one of the plurality of data values is computed for each time series data sequence;
determining an anomaly score for each one of the time series data sequences using the determined numeric representation for each one of the time series data sequences; and
determining a distribution of the determined anomaly scores under normal conditions.
2 . (canceled)
3 . (canceled)
4 . The method of claim 1 , wherein the vector of statistics includes one of a maximum value, a minimum value, a mean value, a standard deviation value, or higher order moments.
5 . The method of claim 1 , wherein the step of determining an anomaly score using the determined numeric representation for each one of the time series data sequences comprises:
determining an average distance in terms of sensed data from each sensor to each one of the plurality of sensors using the determined numeric representation for each one of the time series data sequences;
determining a minimum distance between each one of the plurality of sensors using the determined numeric representation for each one of the time series data sequences; or
determining a sum of the differences between the plurality of sensors using the determined numeric representation for each one of the time series data sequences.
6 . The method of claim 1 , wherein the step of determining a distribution of the determined anomaly scores under normal conditions comprises:
determining a mean and a standard deviation of the determined anomaly scores and applying and applying statistical tests to determine whether or not the determined anomaly scores are within range or are out of range; or
establishing a ranking of the sensors based on the determined anomaly scores and reporting a violation of the established ranking.
7 . The method of claim 1 , wherein the plurality of correlated sensors comprise temperature sensors located within a defined area.
8 . A system for detecting an anomaly in data provided by each one of a plurality of correlated sensors includes a processor in communication with one or more types of memory, the processor being configured to:
receive from each one of the plurality of correlated sensors a corresponding time series data sequence, each data sequence representing a plurality of data values sensed by a corresponding one of the plurality of correlated sensors at a sampling frequency, each of the data values of each data sequence being sensed at a particular point in time in the time series data sequence, wherein the correlated sensors have a correlation related to a common measured parameter or a relative proximity to one another;
determine a numeric representation for each one of the time series data sequences, wherein on a condition that the sampling frequency is not the same for each one of the plurality of correlated sensors, a vector of statistics for each one of the plurality of data values is computed for each time series data sequence;
determine an anomaly score for each one of the time series data sequences using the determined numeric representation for each one of the time series data sequences; and
determine a distribution of the determined anomaly scores under normal conditions.
9 . (canceled)
10 . (canceled)
11 . The system of claim 8 , wherein the vector of statistics is one of a maximum value, a minimum value, a mean value, a standard deviation value, or higher order moments.
12 . The system of claim 8 , wherein when the processor determines an anomaly score using the determined numeric representation for each one of the time series data sequences, the processor further:
determines an average distance in terms of sensed data from each sensor to each one of the plurality of sensors using the determined numeric representation for each one of the time series data sequences;
determines a minimum distance between each one of the plurality of sensors using the determined numeric representation for each one of the time series data sequences; or
determines a sum of the differences between the plurality of sensors using the determined numeric representation for each one of the time series data sequences.
13 . The system of claim 8 , wherein when the processor determines a distribution of the determined anomaly scores under normal conditions, the processor further:
determines a mean and a standard deviation of the determined anomaly scores and applies statistical tests to determine whether or not the determined anomaly scores are within range or are out of range; or
establishes a ranking of the sensors based on the determined anomaly scores and reports a violation of the established ranking.
14 . The system of claim 8 , wherein the plurality of correlated sensors comprise temperature sensors located within a defined area.
15 . A computer program product for detecting an anomaly in data provided by each one of a plurality of correlated sensors comprises a computer readable storage medium having computer executable instructions embodied thereon, the computer readable storage medium comprises instructions to:
receive from each one of the plurality of correlated sensors a corresponding time series data sequence, each data sequence representing a plurality of data values sensed by a corresponding one of the plurality of correlated sensors at a sampling frequency, each of the data values of each data sequence being sensed at a particular point in time in the time series data sequence, wherein the correlated sensors have a correlation related to a common measured parameter or a relative proximity to one another;
determine a numeric representation for each one of the time series data sequences, wherein on a condition that the sampling frequency is not the same for each one of the plurality of correlated sensors, a vector of statistics for each one of the plurality of data values is computed for each time series data sequence;
determine an anomaly score for each one of the time series data sequences using the determined numeric representation for each one of the time series data sequences; and
determine a distribution of the determined anomaly scores under normal conditions.
16 . (canceled)
17 . (canceled)
18 . The computer program product of claim 15 , wherein the vector of statistics includes one of a maximum value, a minimum value, a mean value, a standard deviation value, or higher order moments.
19 . The computer program product of claim 15 , wherein when an anomaly score is determined using the determined numeric representation for each one of the time series data sequences, the computer readable storage medium further comprises instructions to:
determine an average distance in terms of sensed data from each sensor to each one of the plurality of sensors using the determined numeric representation for each one of the time series data sequences;
determine a minimum distance between each one of the plurality of sensors using the determined numeric representation for each one of the time series data sequences; or
determine a sum of the differences between the plurality of sensors using the determined numeric representation for each one of the time series data sequences.
20 . The computer program product of claim 15 , wherein when a distribution of the determined anomaly scores is determined under normal conditions, the computer readable storage medium further comprises instructions to:
determine a mean and a standard deviation of the determined anomaly scores and apply statistical tests to determine whether the determined anomaly scores are within range or out of range; or
establish a ranking of the sensors based on the determined anomaly scores and report a violation of the established ranking.
21 . The method of claim 6 , further comprising:
setting thresholds for any deviations of the determined anomaly scores, wherein the deviation may be set manually based on domain experience.
22 . The system of claim 13 , wherein the processor further:
sets thresholds for any deviations of the determined anomaly scores, wherein the deviation may be set manually based on domain experience.
23 . The computer program product of claim 15 , wherein the computer readable storage medium further comprises instructions to:
set thresholds for any deviations of the determined anomaly scores, wherein the deviation may be set manually based on domain experience.