As well as the impacts of LCTs on the characteristics at the household and low voltage level, large uptakes of unknown quantities of distributed generation are already having an impact on planning and assessment tools at the Transmission level. One of the major sources of load forecast errors for the National Grid occurs on sunny days providing large amounts of PV generation at largely indeterminate locations on the network. Linking information from smart metering and energy end usage analytics (WP2), this work package aims to use advanced mathematical and statistical techniques to gain greater visibilities and insights into the demands/generation on the entire grid network from distribution up to the transmission level. This package aims to use state-of-the-art techniques to generate probabilistic forecasts, quantify uncertainty, understand the main drivers of forecasting errors, and characterise network aggregations points such as substations. This will enable operators to infer potential network flexibility and the viability of particular network solutions (such as DSR and storage devices). This will be investigated through the three subtasks:
Hierarchical Probabilistic Load Forecasting – the volatility and increased uncertainty of demand and generation means probabilistic load forecasts are required at all levels of the electrical network hierarchy from household [Haben16] to the transmission level. These forecasts will be augmented with probabilistic demand/generation forecasts of significant low carbon technologies, in particular with a focus on distributed generation. This generates scenario forecasts that will assist network planning and management as well as calibrating the forecasts to estimate current generation resources, which assists in reducing forecast errors at the transmission level. This complements (EP/R023484/1, Browell Fellowship) which is addresses the need for similar hierarchical forecasts but from GSP up to regional and national transmission system level.
Proper Scoring for Characterisation and Reporting Forecast Errors – increases in unknown photovoltaics installations have already introduced significant sources of errors in forecasts at the transmission level. Conventional error metrics have been shown to miss these [Haben14]. In this subtask we use our hierarchical probabilistic forecast framework to identify sources and drivers of forecast errors and uncertainty throughout a network which may be purposed as a virtual balancing unit. This will include the corroboration of weather effects, technology pervasiveness, and customer aggregations.
Substation Flexibility: Drivers and Influencers – through the hierarchical forecasting framework developed in earlier activities and the identification of forecast errors we can then develop a characterisation of the networks. The methods and techniques will be developed using our experience in probabilistic load forecasting, unsupervised discrimination, Gaussian processes, and feature extraction. Inferences can be scaled up from the preliminary data sets to the national level, e.g. through bootstrapping methodologies.