Jethro Dowell

Very-Short-Term Probabilistic Wind Power Forecasts by Sparse Vector Autoregression

A spatio-temporal method for producing very-short-term parametric probabilistic wind power forecasts at a large number of locations is presented. Smart grids containing tens, or hundreds, of wind generators require skilled very-short-term forecasts to operate effectively and spatial information is highly desirable. In addition, probabilistic forecasts are widely regarded as necessary for optimal power system management as they quantify the uncertainty associated with point forecasts. Here, we work within a parametric framework based on the logit-normal distribution and forecast its parameters. The location parameter for multiple wind farms is modeled as a vector-valued spatio-temporal process and the scale parameter is tracked by modified exponential smoothing. A state-of-the-art technique for fitting sparse vector autoregressive models is employed to model the location parameter and demonstrates numerical advantages over conventional vector autoregressive models. The proposed method is tested on a dataset of 5 min mean wind power generation at 22 wind farms in Australia. Five-min ahead forecasts are produced and evaluated in terms of point, and probabilistic forecast skill scores and calibration. Conventional autoregressive and vector autoregressive models serve as benchmarks.

Kernel methods for short-term spatio-temporal wind prediction

Two nonlinear methods for producing short-term spatio-temporal wind speed forecast are presented. From the relatively new class of kernel methods, a kernel least mean squares algorithm and kernel recursive least squares algorithm are introduced and used to produce 1 to 6 hour-ahead predictions of wind speed at six locations in the Netherlands. The performance of the proposed methods are compared to their linear equivalents, as well as the autoregressive, vector autoregressive and persistence time series models. The kernel recursive least squares algorithm is shown to offer significant improvement over all benchmarks, particularly for longer forecast horizons. Both proposed algorithms exhibit desirable numerical properties and are ripe for further development.

A widely linear multichannel wiener filter for wind prediction

The desire to improve short-term predictions of wind speed and direction has motivated the development of a spatial covariance-based predictor in a complex valued multichannel structure. Wind speed and direction are modelled as the magnitude and phase of complex time series and measurements from multiple geographic locations are embedded in a complex vector which is then used as input to a multichannel Wiener prediction filter. Building on a C-linear cyclo-stationary predictor, a new widely linear filter is developed and tested on hourly mean wind speed and direction measurements made at 13 locations in the UK over 6 years. The new predictor shows a reduction in mean squared error at all locations. Furthermore it is found that the scale of that reduction strongly depends on conditions local to the measurement site.

Spatio-temporal prediction of wind speed and direction by continuous directional regime

This paper proposes a statistical method for 1-6 hour-ahead prediction of hourly mean wind speed and direction to better forecast the power produced by wind turbines, an increasingly important component of power system operation. The wind speed and direction are modelled via the magnitude and phase of a complex vector containing measurements from multiple geographic locations. The predictor is derived from the spatio-temporal covariance which is estimated at regular time intervals from a subset of the available training data, the wind direction of which lies within a sliding range of angles centred on the most recent measurement of wind direction. This is a generalisation of regime-switching type approaches which train separate predictors for a few fixed regimes. The new predictor is tested on the Hydra dataset of wind across the Netherlands and compared to persistence and a cyclo-stationary Wiener filter, a state-of-the-art spatial predictor of wind speed and direction. Results show that the proposed technique is able to predict the wind vector more accurately than these benchmarks on dataset containing 4 to 27 sites, with greater accuracy for larger datasets.

A Review of probabilistic methods for defining reserve requirements

In this paper we examine potential improvements in how load and generation forecast uncertainty is captured when setting reserve levels in power systems with significant renewable generation penetration and discuss the merit of proposed new methods in this area. One important difference between methods is whether reserves are defined based on the marginal distribution of forecast errors, as calculated from historic data, or whether the conditional distribution, specific to the time at which reserves are being scheduled, is used. This paper is a review of published current practice in markets which are at the leading edge of this problem, summarizing their experiences, and aligning it with academic modeling work. We conclude that the ultimate goal for all markets expected to manage high levels of renewable generation should be a reserve setting mechanism which utilizes the best understanding of meteorological uncertainties combined with traditional models of uncertainty arising from forced outages.