Projects
Projects: Projects for Investigator |
||
| Reference Number | NIA2_NGESO003 | |
| Title | Probabilistic Machine Learning Solution for Dynamic Reserve Setting | |
| Status | Completed | |
| Energy Categories | Other Cross-Cutting Technologies or Research(Energy Models) 70%; Other Power and Storage Technologies(Electricity transmission and distribution) 30%; |
|
| Research Types | Applied Research and Development 100% | |
| Science and Technology Fields | PHYSICAL SCIENCES AND MATHEMATICS (Computer Science and Informatics) 50%; ENGINEERING AND TECHNOLOGY (Electrical and Electronic Engineering) 50%; |
|
| UKERC Cross Cutting Characterisation | Systems Analysis related to energy R&D (Energy modelling) 100% | |
| Principal Investigator |
Project Contact No email address given National Grid ESO |
|
| Award Type | Network Innovation Allowance | |
| Funding Source | Ofgem | |
| Start Date | 01 May 2021 | |
| End Date | 30 April 2022 | |
| Duration | ENA months | |
| Total Grant Value | £400,000 | |
| Industrial Sectors | Power | |
| Region | London | |
| Programme | Network Innovation Allowance | |
| Investigators | Principal Investigator | Project Contact , National Grid ESO (100.000%) |
| Other Investigator | Project Contact , National Grid ESO (100.000%) |
|
| Industrial Collaborator | Project Contact , National Grid plc (0.000%) |
|
| Web Site | https://smarter.energynetworks.org/projects/NIA2_NGESO003 |
|
| Objectives | This project will look to undertake the following scope of work: Phase 1: Phase 1 will assess the technical feasibility of end-to-end capability development and of the ML solution in terms of data availability and quality. Produce end to end proof of concept (PoC) Demo 1 on Smith Institute platform to identify and agree dependencies on data, systems, and expertise. Breakpoint 1: Identify if data availability or quality issues are insurmountable. Phase 2: Phase 2 will establish a proof of concept ML model and compare its performance with the current reserve setting process. Develop PoC Demo 2 to implement, test, and document the data analysis and machine learning required in PoC Demo. Breakpoint 2: A performance comparison between BAU and new approach. The PoC solution outputs will be compared to the BAU approach to demonstrate superior performance in terms of matching reserve levels to NGESOs risk appetite and the expected cost or risk management benefits.Parallel running of BAU approach and innovative approach for a period of time to build trust and understanding in the ENCC and allowing a live test of cost savings & risk identification and management.Phase 3 (possibly for a future project): Explore options to integrate into control room systems and present a dashboard explaining reserve breakdown, for a period to get the most accurate estimate of cost-savings and build trust in solution. Risk Assessment In line with the ENAs ENIP document, the risk rating is scored Low.TRL Steps = 2 (3 TRL steps)Cost = 1 (£400k)Suppliers = 1 (1 supplier)Data Assumptions = 2 (data will be tested for ML applicability) Dynamic Reserve Level Setting ApproachThe different types of reserve setting processes should be integrated to ensure that the total reserve held is representative of NGESOs risk appetite.In this project, we expect to apply the approach summarised below to both the basic reserve (possibly including interconnectors) and the reserve for renewable energy sources and we expect that our focus will be on 4–24-hour lead times. Dynamic Reserve Setting Solution DesignThis project will create a proof of concept for a DRS solution with the following features:Data pipeline which automates the extraction, cleaning, and preparation of raw data into storage.Probabilistic ML (machine learning) model which makes use of predictor variables in that data (e.g. temperature and wind forecast quantiles, and generation mix) to create more accurate predictions (and prediction intervals) of forecast errors and therefore set reserve levels, which better reflect NGESOs risk appetite.A database and dashboard for the display of the results of the ML model.Automated upload of the ML model results to control room systems.Retraining of the ML model to enable a cycle of continuous learning using new data about recent system conditions and forecast errors. Probabilistic ML model which makes use of predictor variables in that data (e.g. temperature and wind forecast quantiles, and generation mix) to create more accurate predictions (and prediction intervals) of forecast errors and therefore set reserve levels, which better reflect NGESOs risk appetite.Retraining of the ML model to enable a cycle of continuous learning using new data about recent system conditions and forecast errors. | |
| Abstract | Currently, reserve levels are based on statistical analysis of historical generation and forecasting errors. Using artifical intelligence and machine learning this project will look to set reserve levels dynamically, day ahead. When considering additional interconnectors and increasing weather driven effects on the uncertainty of visible demand (impacted by unmetered embedded generation) and metered generation, we would anticipate further increases in reserve holdings without innovative models and approaches to the challenges presented by a changing system. This innovation project attempts to prototype an advanced ML model as an ambitious methodology to capture the various sources of uncertainty, to be created in parallel with the BAU development of new models.These algorithms would forecast reserve requirements by learning from historical behaviour and input drivers to find correlations with the uncertainty of system conditions. | |
| Publications | (none) |
|
| Final Report | (none) |
|
| Added to Database | 19/10/22 | |
