Learnings

Outcomes

Outcomes from the project are:

·       Model Outputs: Predicted transformer load for both ground mounted and pole mounted transformer model ensuring 100% visibility of the LV network for all three UK Power Networks’ areas (LPN, SPN and EPN).

·       Granular view: The outputs of the model offer a detailed perspective, presenting yearly, monthly and half hourly values for transformer load throughout all three network areas.

·       Appropriate investment plan: The outputs of the RTU implementation tool will recognise priority regions for the installation of hardware monitoring devices, enabling the implementation of a suitable investment plan.

·       Business efficiency: Granular view of the model output combined with output of RTU implementation tool offers enhanced business efficiency through improved planning decisions, optimised LV monitoring and utilisation band intervention.

·       Established data pipelines: Data management framework including data architecture and Azure data pipeline infrastructure, is established for future projects, including innovation, to adhere to.

·       Improved data governance: Data quality governance process has evolved to enhance the effectiveness of resolving data quality problems and resolution at source systems, ultimately benefitting all users of the system.

·       Flexibility procurement: The predicted load values facilitate more focused procurement of flexibility, particularly as unmonitored sites that would have otherwise might have been excluded.

·       Profiled connections: A view on profiled connections and related solutions with a focus on understanding optimum number of time periods depending on site constraints and overall EV load forecasts for fleet operators.

Lessons Learnt

The key lessons learnt from WP1 and WP3 are described below:

·       Data quality is critical: Data models are only as good as the data they're trained on. Therefore, it is essential to ensure that the data is accurate, reliable, and free of bias. Data preparation and cleaning are critical steps that cannot be overlooked. It is essential to have a data catalogue and data governance in place to resolve data quality issues efficiently.

o   Key data quality issues addressed during the project that have a high impact on model accuracy included mismatches in customer connectivity/allocation and missing or mismatched transformer location and postcode data. Prediction accuracy between network areas and transformer sizes was also influenced by the extent of real-time monitoring data availability.

·       Delta data: Defining the frequency and delta dataset refresh is of utmost importance and a critical element for the solution's future, It not only improves the model and data quality but also aids in reducing Azure costs necessary for the data transfer process. 

·       Focus on business outcomes and validation approaches: It is crucial to keep the business objectives in mind this could include model method statements, model output data and the frequency of refresh of data. A clear understanding of the validation approach is critical. Once established it should be documented and implemented for each model.

·       Collaboration is key: This project required cross-functional collaboration between data scientists, subject matter experts, IT professionals, and business stakeholders. Effective communication and collaboration are necessary to ensure that everyone is aligned and working towards the same goals.

·       Efficient time management: A significant amount of time and resources is essential. Such projects often involve a lot of trial and error. Therefore, it is important to set realistic expectations and timelines.

·       Continuous learning and improvement: Data models require constant updates and improvements to remain effective. It is essential to have a process in place for monitoring the model's performance and making necessary adjustments. It is crucial to regularly track progress at specified intervals, such as every month, and comprehend the reasons behind any variations in forecasted accuracy.

·       Ethical considerations: Data models should be developed with ethical considerations in mind. Bias, privacy, and transparency are critical issues that need to be addressed. It is important to ensure that models are designed and developed in an ethical and responsible manner. Make sure that a mutually agreed test-train split is upheld, and that the data models undergo cross validations. 

·       Technical feasibility: Before starting the project, it is essential to assess its technical feasibility. The availability of data, resources, and expertise is critical. It is important to evaluate whether the necessary technology and infrastructure namely data bricks, azure data pipeline and central data store are in place to support the project.

The key lessons learnt from WP2 are described below:

·       Most potential external data providers engaged showed readiness and interest in sharing customer LCT and DER data. The ones that didn't were held a limited amount of assets or a limited amount of data to consider data sharing as a potential commercial opportunity in their roadmaps.

·       Maturity levels of different data providers varied vastly in terms of business and technology readiness. Diverse understanding of data sharing policies, frameworks, and contracts, and managing customer consent for data sharing, would make it difficult to standardise incentives and data sharing agreements.  

·       The knock on effect of finding it difficult to quantify the business value of external LCT and DER data is that it is difficult to quantify the price a network operator is willing to pay for the services necessary to procure this data and for the data itself.

·       Bigger organisations shared a degree of scepticism for dealing with innovation projects in this space – specifically in innovation projects where the sustainability of the project is uncertain (i.e. the gap between innovation and BAU).