A Bi-LSTM Autoencoder Framework for Anomaly Detection -- A Case Study of a Wind Power Dataset