Weather Station Selection for Electric Load Forecasting

In the load forecasting literature, most papers are focusing on the direct application of some techniques, such as regression, ARIMA, ANN, etc. Not many papers are discussing original methodologies that can be used across different techniques. The investigation on new sub-problems of load forecasting is even rare. Weather station selection is a necessary step in load forecasting, but has never been formally studied in the past many decades. We wrote this paper last year to describe how to apply a greedy method and out-of-sample test to select weather stations. Although regression models are used here, our methodology is independent of the techniques as long as they rely on weather variables.

The paper was accepted by International Journal of Forecasting this summer. The working paper is available HERE. I will update the citation once the paper is on Science Direct.

Citation
Tao Hong, Pu Wang and Laura White, "Weather station selection for electric load forecasting", International Journal of Forecasting, vol.31, no.2, pp 286-295, April-June, 2015, working paper available from http://www.drhongtao.com/articles.

Weather Station Selection for Electric Load Forecasting

Tao Hong, Pu Wang and Laura White

Abstract

Weather is a major driving factor of electricity demand. Selection of weather station(s) plays a vital role in electric load forecasting. Nevertheless, minimal research efforts have been devoted to weather station selection. In the smart grid era, hierarchical load forecasting, which provides load forecasts throughout the utility system hierarchy, is becoming an emerging and important topic. Since there are many nodes to forecast in the hierarchy, it is no longer feasible for forecasting analysts to manually figure out the best weather stations for each node. A commonly used solution framework is to assign the same number of weather stations to all nodes at the same level of the hierarchy. This framework was also adopted by all of the four winning teams of Global Energy Forecasting Competition 2012 (GEFCom2012) in the hierarchical load forecasting track. In this paper, we propose a weather station selection framework to determine how many and which weather stations to use for a territory of interest. We also present a practical, transparent and reproducible implementation of the proposed framework. We demonstrate the application of the proposed approach to forecasting electricity at different levels in the hierarchies of two US utilities respectively. One of them is a large US generation and transmission cooperative that has deployed the proposed framework. The other one is from GEFCom2012. In both case studies, we compare our unconstrained approach with four other alternatives based on the common practice mentioned above. We show that the forecasting accuracy can be improved by releasing the constraint on the fixed number of weather stations.