Communication Network Predictive Models for Drift Evaluation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Complex communication networks face challenges with predictive models due to model drift, making it unclear if predicted failures would have occurred without corrective actions, thus requiring periodic re-training and re-evaluation.

Innovation Solution

A computer-implemented method for managing predictive models in communication networks involves allocating network devices into first and second clusters, disabling the predictive model in the first cluster, enabling it in the second cluster, collecting data from both clusters, and analyzing it to detect deviations, allowing for concurrent use and re-training of the model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the predictive model is periodically re-trained and re-evaluated, then model accuracy is maintained, but network service may be disrupted during re-training

Engineering Contradiction:

Improvepredictive model accuracyVSAvoidnetwork service continuity

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

Data is collected from network devices in advance and stored for later model evaluation. This preliminary data collection allows the predictive model to be re-trained and re-evaluated using historical data without requiring real-time network interruption, thus maintaining network service continuity while ensuring model accuracy.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

A data collection and storage mechanism acts as an intermediary between the predictive model and the network devices. This intermediary layer allows the model to be re-trained using collected data without directly interrupting network operations, enabling model maintenance while preserving service continuity.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Quantity of substance

If data is collected from all network devices continuously, then sufficient data is available for model analysis, but data storage and processing requirements increase

Engineering Contradiction:

Improvedata quantity for model analysisVSAvoiddata management complexity

Core Design Contradiction:

Quantity of substanceVSDevice complexity

Solution Approach 1:

Network devices are segmented into first and second clusters for differential data collection purposes. Data is collected from both clusters, but the predictive model is only enabled in the second cluster. This segmentation optimizes data collection by focusing model application on one cluster while still gathering data from both, reducing unnecessary processing overhead while maintaining sufficient data quantity for model analysis.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS12284086B2Management of predictive models of a communication network

Publication Date: 2025.04.22 ELISA OYJ

AI summary

A computer implemented method of managing a predictive model of a communication network, wherein the predictive model is configured to identify and correct forthcoming failures in network devices based on data collected from the network devices. The method includes allocating network devices of the communication network into first and second clusters, disabling the predictive model in the first cluster, enabling the predictive model in the second cluster, collecting data from the first cluster, repeating said disabling, enabling and collecting with a new allocation of first and second clusters to continuously collect data with different allocation of first and second clusters, and outputting at least the data collected from the first cluster for analysis of the predictive model.