ML.ai

Automated process of finding useful patterns in data, these patterns may be saved as models to make predictions and explain results. Process is ideal for existing data stores including PODS GIS, in-line inspections, leak surveys, corrosion surveys, domain expert observations, any structured data or any results from deterministic structures such as existing risk algorithms, code rules or engineering formulas.

Distribution Pipelines - fully compliant data driven qualitative or quantitative risk assessment based on machine learning processes and advanced analytics. Models are learned based on client specific observations augmented by industry available data and public data stores.  Dynamic segmentation considers typical distribution network configurations.

Machine learned results of any prediction are fully explainable through breakdown and shapley analysis. Explainability supports mitigation decision-making, resource allocation, spend prioritization and practical discussion of data driven results.

The simulator allows the practitioner to modify inputs to output new predictions and assess model behavior in high dimensional space.

Data driven machine learned models are validated based on in-line inspections, visual inspections, corrosion surveys or domain expert interviews. This feature supports PHMSA requirements relating to improving model performance and transparency.

Transmission Pipelines - fully compliant data driven qualitative or quantitative risk assessment based on machine learning processes and advanced analytics. Models are learned based on client specific observations augmented by industry available data and public data stores. Dynamic segmentation considers typical transmission system configurations.

Facilities - software supports unstationed asset analysis for facility risk. Compressors, regulator stations, valves, tanks, etc. may be managed through the machine learning process to calculate qualitative or quantitative risk. 

Models are fully explainable through predictor importance values  supporting mitigation decision-making, resource allocation and spend prioritization. 

Unsupervised learning supports rare threat analysis and susceptibility classification modeling. Complex high dimensional data structures are represented in 2 dimensional space to reveal useful risk patterns.

PLR's machine learning process accommodates multiple methods with tools available to find the best model based on performance objectives.