Font Size: 

Climate change is an issue caused in no small part by nonrenewable
energy generation [1] and low efficiency in electric systems [2]. A possible solution involves switching to renewable energy systems e.g. solar, wind, hydro, etc [2]. However, this presents a different problem that of intermittent power production [3]. Since the energy output cannot be controlled at any given time, a feasible strategy involves modifying consumption to fit into the supply, i.e. demand-side management (DSM). In order to use DSM strategies it is important to determine the possibility of demand-side flexibility, which is a measure of how demand can be modified to fit the supply [3]. To calculate flexibility it is necessary to first identify patterns within a power demand time series. This study presents a pattern recognition method developed for Mexican power demand time series, consisting of prepossessing, segmentation, dynamic time warping and clustering. Note that the data used is below the average Mexican household consumption rate of 4.76KWh [4], thus power demand can be volatile and present anomalies that can affect pattern segmentation [5]. The proposed solution involves using a triangular moving average to smooth the data. The next step is to separate the time series into possible patterns by analyzing rises and falls within the power for determined periods of time. Subsequently the distance between every sequence stored is calculated by using dynamic time warping, which
can measure segments of different lengths in order to define relative closeness in shape [6]. By creating a distance matrix of every measure between two segments, a clustering algorithm can be applied, i.e. affinity propagation. This algorithm uses a similarity matrix as input and outputs the data grouped by closeness and defines centers for each group [7]. In the present use-case several patterns have been identified.

Pattern Recognition, power Time Series, Clustering