Conventional approaches to dynamic line rating (DLR) forecasting provide single point estimates with no indication of the distribution of possible errors. Furthermore, most research related to DLR forecasting deals only with continuous or steady-state ratings while less attention has been given to short-term or transient-state ratings. This thesis describes (a) weather-based models to estimate probabilistic forecasts of steady-state DLRs for up to three 10-minutes time steps ahead for a particular span and a complete overhead line (OHL) and also (b) a fast-computational weather-based approach to probabilistic forecasting of transient-state DLRs for a particular span for time horizons of 10, 20 and 30 minutes. The percentiles of DLR forecasts can be used by a system operator within a chosen risk policy informed by the probability of a rating being exceeded. The thesis first develops time series forecasting models for different weather variables that impact on line rating (i.e. air temperature, wind speed, wind direction and solar radiation) at weather stations that are installed along the route of 132kV OHLs in North Wales. Predictive centres of weather variables are modelled as a sum of residuals predicted by a suitable auto-regressive process and temporal trends fitted by Fourier series. Conditional heteroscedasticity of the predictive distribution is modelled as a linear function of recent changes in residuals within one hour for air temperature and wind speed or concentration of wind direction observations within the most recent two hours. A technique of minimum continuous ranked probability score estimation is employed to determine predictive distributions of the measured weather variables. Then the thesis uses Monte Carlo simulation to generate a large number of random weather samples from the modelled predictive distributions which are paired to have rank correlations similar to those among their recent observations. The probabilistic steady-state DLR forecasts for a particular span in proximity to a weather station are estimated from the random weather samples combined with a maximum allowable conductor temperature using a thermal model of the conductors (i.e. a steady-state heat balance equation). For a complete OHL, possible weather predictions at each span are inferred from random weather samples at stations by using suitable spatial interpolation models; the steady-state DLR forecast of the OHL is then identified as the minimum DLR among all spans for each generated scenario. Using an enhanced analytical method which evolves from a non-steady-state heat balance equation to track the transient-state conductor temperature, the transient-state DLR forecast for a particular span is calculated as that which increases the conductor temperature from an initial value to the maximum allowable limit for a particular future time period (i.e. in this study, 10, 20 and 30 minutes) under each set of random weather samples. The calibration of probabilistic DLR forecasts estimated from independent or correlated random weather samples are then examined to determine which approaches are most suited to estimation of DLRs at the lower end of a predictive distribution consistent with a system operator’s risk policy. The potential use of DLR forecasting is then evaluated through estimating the degree to which wind generation curtailment for various assumed installed capacities at a wind farm that is connected to the 132kV network in North Wales can be alleviated through using the lower percentiles of steady-state DLR forecasts in place of the SLRs for each 132kV OHL.
|Award date||6 Nov 2018|
|Place of Publication||Glasgow|
|Publication status||Published - 6 Nov 2018|
- DLR forecasting
- dynamic line rating forecasting
- fast-computational weather-based approach
- weather variables