Digital Tribes Houston16 Feb - 23 Feb
A high intensity hackathon testing digital skills against resource industry challenges to identify global talent.
This competition has finished.
Automatic Water Oil Separator Upset Detection
There are many classical time series anomaly detection algorithms. What we used is called Seasonal Trend Loess decomposition, which can decompose a time series into three parts: seasonal, trend, residual. A large value in the residual can be considered as an anomaly. Using this algorithm, we created a label for each timestamp to indicate whether it is an anomaly or not. Then We feed selected features along with the created label into a Long Short Term Memory (LSTM) model, which is one typical structure of recurrent neural network.