دانلود رایگان مقاله همبستگی و خوشه بندی در بازارهای برق عمده فروشی

Abstract

We study the structure of locational marginal prices in day-ahead and realtime wholesale electricity markets. In particular, we consider the case of two North American markets and show that the price correlations contain information on the locational structure of the grid. We study various clustering methods and introduce a type of correlation function based on event synchronization for spiky time series, and another based on string correlations of location names provided by the markets. This allows us to reconstruct aspects of the locational structure of the grid.

7. Conclusions

In this paper we discussed various correlation, filtering, and clustering methods with the purpose of analyzing the underlying structure of wholesale electricity market data. Specifically, we have analyzed the nodal price difference between day-ahead and real-time, which are relevant for understanding the inefficiencies in the planning process that outputs the day-ahead prices. We also introduced an alternative measure of correlation based on a modification of Event Synchronization, which focuses on more rare and strong events (spikes) in time series. We have argued that most of the correlations observed in our dataset are due to random fluctuations and that these should be filtered, or a nonlinear measure of correlation should be used instead. In addition to this, we explored various methods for inferring the correlation structure for MISO, identifying various clusters. These methods are robust as they use only the strongest correlation links.

Finally and most importantly, we provided a framework to choose and filter correlation matrices of electricity price data by using the string kernel correlation proxy, which we believe can be used to test the validity of our methodology.

We also extended our results to the time domain and analyzed how the market structure changes from week to week. This analysis provides a quantitative method to infer non-stationarity as well as to test the clustering methods. We concluded that the MST modularity maximization method is more suitable than spectral clustering. Furthermore, even after filtering multiple types of noise, the clustering results for RMT filtered correlation matrices were found to be, in some cases, unstable over time (with respect to the proxy), although it has achieved the best performance. Of all the methods, we have found Pearson based and the Event Synchronization methods to be the most stable overtime. Overall, we have found Event Synchronization in conjunction with MST to be the most appropriate clustering method for its stability and remarkable performance. An interesting avenue for future work is to try to reconstruct the network using methods of graph inference based on entropy maximization.