Anomaly detection using state‐space models and reinforcement learning

Khazaeli, Shervin; Nguyen, Luong Ha; Goulet, James A.

doi:10.1002/stc.2720

Cited by 10 publications

(2 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Because the state-space model can determine the value excluding noise data via smoothing, it is considered an effective method, especially when evaluating past diseases. In fact, the state-space model is also used for change-point and anomaly detection [16,17]. However, there are few reports on the construction of predictive models using the state-space model in the veterinary eld; it is a bit more di cult to build predictive models using this model than other predictive models.…”

Section: Discussionmentioning

confidence: 99%

Evaluating swine disease occurrence on farms using the state-space model based on meat inspection data: a time-series analysis

Narita,

Kubo,

Nagakura

et al. 2023

Preprint

View full text Add to dashboard Cite

Background Data on abnormal health conditions in animals obtained from slaughter inspection are important for identifying problems in fattening management. However, methods to objectively evaluate diseases on farms using inspection data has not yet been well established. It is important to assess fattening management on farms using data obtained from slaughter inspection. In this study, we developed the state-space model to evaluate swine morbidity using slaughter inspection data. Results The most appropriate model for each disease was constructed using the state-space model. Data on 11 diseases in slaughterhouses over the past 4 years were used to build the model. The model was validated using data from 14 farms. The local-level model (the simplest model) was the best model for all diseases. We found that the analysis of slaughter data using the state-space model could construct a model with greater accuracy and flexibility than the ARIMA model. In this study, no seasonality or trend model was selected for any disease. It is thought that models with seasonality were not selected because diseases in swine shipped to slaughterhouses were the result of illness at some point during the 6-month fattening period between birth and shipment. Conclusion Evaluation of previous diseases helps with the objective understanding of problems in fattening management. We believe that clarifying how farms manage fattening of their pigs will lead to improved farm profits. In that respect, it is important to use slaughterhouse data for fattening evaluation, and it is extremely useful to use mathematical models for slaughterhouse data. However, in this research, the model was constructed on the assumption of normality and linearity. In the future, we believe that we can build a more accurate model by considering models that assume non-normality and non-linearity.

show abstract

Section: Discussionmentioning

confidence: 99%

Evaluating swine disease occurrence on farms using the state-space model based on meat inspection data: a time-series analysis

Narita,

Kubo,

Nagakura

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The policy-based anomaly detector outperformed the valuebased approach. Khanzaeli applied a valued-based anomaly detector with state-space models to structural health monitoring, where Q-learning was used, and artificial anomalies were generated for test purposes [53].…”

Section: Deep Reinforcement Learning-based Fault Detectionmentioning

confidence: 99%

Label-Free Fault Detection Scheme for Inverters of PV Systems: Deep Reinforcement Learning-Based Dynamic Threshold

et al. 2023

View full text Add to dashboard Cite

Generally, photovoltaic (PV) fault detection approaches can be divided into two groups: end-to-end and threshold methods. The end-to-end method typically uses a deep neural network (DNN) to learn fault patterns from labeled datasets, which directly detect whether faults occur or not. The threshold method first estimates power generation and uses thresholds to detect atypical deviations of measured values from estimated ones. The former method heavily relies on fault-labeled data and, therefore, requires the collection of abnormal event records, which is usually difficult, due to the sparseness of these events. The latter method typically uses statistical approaches, such as 3-sigma, to find thresholds, and it can be practically utilized without fault labels. However, setting a threshold with a proper confidence interval is still challenging, as PV power generation is sensitive to variations in environmental conditions, such as irradiance, ambient temperature, wind speed and humidity. In this paper, we propose a novel deep reinforcement learning (DRL)-based label-free fault detection scheme in which thresholds are dynamically assigned with suitable confidence intervals under varying environmental conditions. Various weather properties were used as input features (i.e., states) to a DRL agent, and proper thresholds were estimated in real time from the actions of the DRL agent. To this end, a reward function was designed for learning proper thresholds without fault labels under different weather conditions. To evaluate the performance of the proposed scheme, the PV dataset of the National Institute of Standards and Technology (NIST) was used, as it includes paired records of local weather and PV generations. The DRL-based scheme was compared with static and conventional dynamic threshold methods, based on statistical approaches. The results revealed that the proposed scheme outperformed the existing methods, providing a 5.67% higher F1-score in the NIST dataset.

show abstract