Figure 1: The state of charge (SoC) of a data logger over time.<\/figcaption><\/figure>\nWe used an SoC threshold to define the event. Let us illustrate with an example. Figure 1 shows the SoC of a data logger over time. The data logger had its batteries changed in March and September. This is inferred from the sudden spike in the SoC. We define an event to have occurred when the SoC of a data logger reaches 30%. Thus, an event occurred in September, as the data logger reached 30% SoC and subsequently had its batteries replaced. In contrast, the data logger had its batteries replaced at 60% SoC in March. The 30% SoC threshold is not observed. This is censoring, where the data logger \u201cdrops out\u201c before experiencing the event.<\/p>\n
<\/span>Labelling Individual Cycles<\/span><\/h3>\nAfter defining the threshold event, we identified the points where a data logger experiences an event or is otherwise censored. To this end, we segmented the SoC curve of each logger into distinct cycles. An event or censorship causes a cycle to end and a new cycle to begin. A cycle can also be defined as the time between battery replacements. Thus, the data logger in Figure 1 has 3 total cycles, 2 of which are censored.<\/p>\n
As the cycles were not labeled in the data, we used a peak detection algorithm from the Python Scipy library to identify the starting points of the individual cycles. The algorithm compares neighboring points in a curve to find local maxima. These local maxima correspond to the aforementioned starting points, which are easily visually identified as sharp spikes in the SoC. By marking these points, individual cycles can be segmented.<\/p>\n