Spatially Enabled Asset Management (SEAM)

Objective(s)

In line with the overall objective of creating and testing a machine learning algorithm to identify and propose fixes for GIS data issues, the project objective are to;

• Generate of potential hypotheses to test and use cases for the tool to be applied to
• Understand the data available to support the machine learning proof of concept
• Outline of the model design including selection of machine learning algorithms.
• Create a final cleaned and prepared dataset that will be used to train and develop the model.
• Provide an interim report that sets out early findings from the modelling and direction for the remainder of the project.
• Develop the final version of the PoC model and front end.
• Carry out statistical evaluation of the model and accuracy through comparison of the model outputs with baseline and training datasets.
• Carry out data cleaning and loading of selected network area, including schematics if available in the format of a connectivity and impedance electrical model of EHV, HV and LV networks.

Provide a summary of key findings, assessment of outcomes against success criteria, recommendations and learnings to be shared.

Problem(s)

Geographic Information Systems for utilities have been created by the digitisation of paper records. These have then undergone transformations as data has been moved between one system and another and manipulations such as the corrections to the Master Map background. It is expected that there will be inaccuracies in the current records and these can persist for many years partly due to the lack of visibility of buried assets and partly due to the length of time between sites requiring site visits that would be expected to update the GIS.

Inaccuracies in the GIS system have the potential to impact:

• Accuracy of network modelling
• Accuracy of regulatory reporting
• Field safety
• Network operational efficiency
• Network upgrade/maintenance efficiency
• Accuracy of New Connections information
• Accuracy of information provided to third parties

At the same time the process to correct identified GIS problems, for example a micro-disconnect issue may be highly manual and therefore costly and time consuming. Therefore, the use of an AI engine to police and correct this essential GIS data is a more effective and efficient method of risk reduction.

Similarly, when performing power systems analysis on the networks, data gaps (such as missing cables and asset types) can be a key issue and can greatly increase the analysis time due to time spent fitting data and cleansing the set prior to carrying out the analysis. With the data having potentially many users within and outside of WPD, every time data gaps are filled this not only duplicates effort but is likely to result in different assumptions being made and potentially different conclusions being drawn. A process that can reduce data gaps consistently and accurately would help resolve these issues.

Method(s)

The SEAM project aims to investigate how Machine Learning (ML) can be employed to carry out data cleansing and data gap closure. The ML model will be trained using an existing dataset and then the ability of the model to successfully identify and correct data issues will be evaluated with a separate dataset that has had errors introduced. This will be followed by application to an unaltered WPD dataset with the results being compared to the errors identified by WPD’s Integrated Network Model.

See our latest webinar below:

• Scope

  The project will include LV and 11kV networks which represent the bulk of GIS data. 33kV networks are included to provide a comparison between the issues identified by the ML algorithm and those flagged by the Integrated Network Model. To enable the use of INM data, the South West region will be used though the approach could be applied to any area.

  The project will investigate model inputs, outputs and different machine learning algorithms with the final model incorporated into a user interface. The design and algorithms used will be documented to enable knowledge transfer.

• Success criteria

  The project will be judged successful if the following criteria are met.

  • A standalone AI Model has been developed tested and applied to a dataset in the agreed regional area.
  • The model performance has been evaluated and the application to the wider GIS data landscape assessed.
  • The approach to roll into business-as-usual has been assessed with recommendations.
  • Key learnings have been identified and shared with other DNOs.
• Potential for new learning

  The learning generated will relate to the core project aims. i.e.

  1. To what degree can machine learning be applied to GIS data to find and fix the more complex errors?
  2. Can the ML approach provide useful indices for confidence in the presence of an error or in the proposed solution?

  It is expected that additional learning will be generated along that way e.g.

  • What types of error are most prevalent in GIS data?
  • What types of error can be corrected using only the GIS data itself?
  • What types of error require additional data from other systems to provide the context for the ML algorithms?
• Documents and links

  Project Registration and PEA Document

  Interim Findings Report

  Model Design Document

  Model Evaluation Report

  Project Dissemination Webinar Slides

  Project Closedown Report - Nov 21

  ENA Closedown Report 

• Monthly updates

  • {{update.title}}

  Show more