Method and system for modeling likely invariants in distributed systems
View Patent ↗Disclosed is a method and system for modeling invariant relationships between flow intensity measurements in a distributed system. In the method, a measurement is randomly selected from a plurality of flow intensity measurements. The method searched for relationships between the randomly selected measurement and each remaining one of the plurality of flow intensity measurements, and each of the flow intensity measurements having a relationship with the randomly selected measurement is grouped into a cluster with the randomly selected measurement. The method than determines relationships between all of the flow intensity measurements in the cluster. This method is repeated with the remaining flow intensity measurements until all of the flow intensity measurements are grouped into a cluster.
1. A method for fault detection in a distributed system comprising:
(a) randomly selecting one measurement from a plurality of flow intensity measurements;
(b) searching for a relationship between said one measurement and each remaining one of the plurality of flow intensity measurements;
(c) grouping said one measurement and each one of the plurality of flow intensity measurements other than said one measurement grouped into the cluster;
(d) determining relationships between each of the flow intensity measurements other than said one measurement grouped into a cluster;
(e) iteratively repeating steps (a)-(d) for remaining ones of the plurality of flow intensity measurements not grouped into a cluster until each of the remaining ones of the plurality of flow intensity measurements is grouped into a cluster;
(f) validating relationships determined between ones of the plurality of flow intensity measurements to determine whether the relationships are invariant relationships;
(g) predicting flow intensities based on the relationships determined to be invariant relationships; and
(h) detecting faults in the distributed system by comparing measured flow intensity measurements to the predicted flow intensities.
2. The method of claim 1 , wherein step (b) comprises:
calculating a model of a relationship between said one measurement and each remaining one of the plurality of flow intensity measurements;
calculating a fitness score for each model; and
comparing the fitness score for each model to a threshold to determine which of the remaining ones of the plurality of flow intensity measurements have a relationship with said one measurement.
3. The method of claim 2 , wherein said step of calculating a model of a relationship between said one measurement and each remaining one of the plurality of flow intensity measurements comprises:
modeling the relationship between said one measurement and each remaining one of the plurality of flow intensity measurements using an AutoRegressive model with eXogenous inputs (ARX).