The behavior of a monitored condition over time for a reactor or reaction system can be analyzed using groupings or windows of data to identify anomalous features in the time-average values. Anomalous features can be identified based on a threshold value generated from the analysis. Based on identification of an anomalous feature, a corrective action can be taken. For example, when the monitored condition is the pressure drop across a catalyst bed, detection of an anomaly can indicate the time to initiate a wash process for the catalyst bed before a large drop in catalyst activity occurs. By detecting an anomaly at an earlier point in time, a wash cycle can be initiated earlier so that the wash is more effective at restoring the catalyst bed to a desired condition.