# US 2025/0086045 A1 · Time series anomaly detection with rare event failure prediction

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0086045 A1
- Document ID: US20250086045A1
- Status: Published application
- Kind: A1
- Application number: 18/466,392
- Filed: 2023-09-13
- Published: 2025-03-13
- Earliest claimed: 2023-09-13
- Assignees: Dell Products L.P.
- Applicants: Dell Products L.P.
- Inventors: Michael Barnes; Sumanta Kashyapi; Niharika Karia; David C. Sydow; Gajanan S Natu
- Web page: https://patentgrep.com/patent/US20250086045A1

## CPC

- G06F 11/079: {Root cause analysis, i.e. error or fault diagnosis (in a hardware test environment G06F11/22; in a software test environment G06F11/36)}
- G06F 11/3419: {by assessing time}
- G06F 11/008: {Reliability or availability analysis}
- G06F 11/0721: {within a central processing unit [CPU]}
- G06F 11/3452: {Performance evaluation by statistical analysis}
- G06F 11/3457: {Performance evaluation by simulation}
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/00: Machine learning
- G06N 20/20: Ensemble learning
- G06F 2201/81: Threshold
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A method for time series anomaly detection with rare event failure prediction includes determining, by a device including a processor and using a first machine learning model, a location of a data anomaly within sequential data. The method also includes determining, by the device and using a second machine learning model that is not the first machine learning model, whether a probability of a future failure event is at least a threshold probability based on the data anomaly and the sequential data.

## Claims

### Claim 1

A system, comprising:
a memory that stores executable components; and
a processor that executes the executable components stored in the memory, wherein the executable components comprise:
an anomaly detection component that locates, using a first machine learning model, anomalous data points within a data sequence; and
an event prediction component that determines, using a second machine learning model that is not the first machine learning model, whether a likelihood of a future service disruption event is at least a threshold likelihood based on the anomalous data points located by the anomaly detection component and the data sequence.

### Claim 2 (depends on 1)

The system of claim 1, wherein the first machine learning model is an unsupervised machine learning model, and wherein the second machine learning model is a supervised machine learning model.

### Claim 3 (depends on 1)

The system of claim 1, wherein the data sequence comprises a stream of time series data indicative of system metrics of a device, the device being selected from a group of devices comprising a computing system device and a data storage device, and wherein the future service disruption event corresponds to a future failure of the device.

### Claim 4 (depends on 3)

The system of claim 3, wherein the second machine learning model is trained using labeled training data corresponding to respective samples of the system metrics preceding previous service disruption events.

### Claim 5 (depends on 3)

The system of claim 3, wherein the system metrics are of a type selected from a group of metric types comprising performance metrics and capacity metrics.

### Claim 6 (depends on 1)

The system of claim 1, wherein the anomaly detection component processes the data sequence using a group of first machine learning models, comprising the first machine learning model, resulting in respective model outputs, and combines the first machine learning models based on the model outputs and a combination criterion.

### Claim 7 (depends on 6)

The system of claim 6, wherein the combination criterion is associated with a voting procedure performed for the first machine learning models.

### Claim 8 (depends on 1)

The system of claim 1, wherein the event prediction component processes the data sequence using a group of second machine learning models, comprising the second machine learning model, resulting in respective model outputs, and selects the second machine learning model from the group of second machine learning models based on the model outputs and a selection criterion.

### Claim 9 (depends on 1)

The system of claim 1, wherein:
the future service disruption event is a first future service disruption event,
the likelihood of the first future service disruption event is a first likelihood,
the threshold likelihood is a first threshold likelihood,
the event prediction component further determines, using a third machine learning model that is not the first machine learning model or the second machine learning model, whether a second likelihood of a second future service disruption event is at least a second threshold likelihood based on the anomalous data points located by the anomaly detection component and the data sequence, and
the first future service disruption event is of a first event type that is not a second event type of the second future service disruption event.

### Claim 10 (depends on 1)

The system of claim 1, wherein the executable components further comprise:
a root cause analysis component that, in response to the event prediction component determining that the likelihood of the future service disruption event is at least the threshold likelihood, determines a predicted cause of the future service disruption event using logical rules.

### Claim 11

A method, comprising:
determining, by a device comprising a processor and using a first machine learning model, a location of a data anomaly within sequential data; and
determining, by the device and using a second machine learning model that is not the first machine learning model, whether a probability of a future failure event is at least a threshold probability based on the data anomaly and the sequential data.

### Claim 12 (depends on 11)

The method of claim 11, wherein the first machine learning model is an unsupervised machine learning model, and wherein the second machine learning model is a supervised machine learning model.

### Claim 13 (depends on 11)

The method of claim 11, wherein the sequential data comprises time series data indicative of metrics associated with a device selected from a group of devices comprising a data storage device and a computing system device, and wherein the future failure event is a system failure event within the device.

### Claim 14 (depends on 13)

The method of claim 13, wherein the second machine learning model is trained using training data corresponding to respective samples of the metrics in respective time intervals preceding past system failure events.

### Claim 15 (depends on 13)

The method of claim 13, wherein the metrics are of a type selected from a group of metric types comprising performance metrics and capacity metrics.

### Claim 16 (depends on 11)

The method of claim 11, wherein the future failure event is a first future failure event, wherein the probability of the first future failure event is a first probability, wherein the threshold probability is a first threshold probability, and wherein the method further comprises:
determining, by the device and using a third machine learning model that is not the first machine learning model or the second machine learning model, whether a second probability of a second future failure event is at least a second threshold probability based on the data anomaly and the sequential data, the first future failure event being of a different event type than the second future failure event.

### Claim 17

A non-transitory machine-readable medium comprising computer executable instructions that, when executed by a processor, facilitate performance of operations, the operations comprising:
determining, using a first machine learning model, a location of an anomalous sample within a data sequence; and
determining, using a second machine learning model that is not the first machine learning model, that a probability of a future disruptive event is at least a threshold probability based on the location of the anomalous sample and the data sequence.

### Claim 18 (depends on 17)

The non-transitory machine-readable medium of claim 17, wherein the first machine learning model is an unsupervised machine learning model, and wherein the second machine learning model is a supervised machine learning model.

### Claim 19 (depends on 17)

The non-transitory machine-readable medium of claim 17, wherein the data sequence comprises a stream of time series data indicative of metrics of a device selected from a group of devices comprising a computing system device and a data storage device, the metrics being selected from a group comprising performance metrics and capacity metrics, and wherein the future disruptive event is a future service disruption within the device.

### Claim 20 (depends on 17)

The non-transitory machine-readable medium of claim 17, wherein the operations further comprise:
in response to the determining that the probability of the future disruptive event is at least the threshold probability, predicting a root cause of the future disruptive event based on a group of logical rules.
