Detecting Abnormal Operations in Concentrated Solar Power Plants from Irregular Sequences of Thermal Images

TL;DR

The paper addresses anomaly detection in irregular sequences of infrared images from CSP plant receivers, where temporal non-stationarity and irregular sampling complicate monitoring. It introduces ForecastAD, a forecasting-based AD model with an image encoder, a context encoder using sinusoidal time embeddings and an LSTM, and an image decoder that forecasts future frames; anomalies are detected via forecast errors, with the anomaly score defined as $s(x,t)=\|x-\hat{x}\|_F^2$. ForecastAD is shown to outperform state-of-the-art baselines across multiple evaluation setups (AUROC and AUPR), especially in scenarios with low-temperature normal samples at cycle starts/ends, and it is successfully deployed on five months of unseen CSP data for maintenance insights. The work provides a reproducible framework, including a labeled dataset subset and a simulated dataset, and demonstrates the practical impact of temporally aware, forecast-based anomaly detection on industrial CSP plant reliability and uptime. Future directions include improving robustness to distribution shifts via probabilistic forecasting and further exploration of context length effects.

Abstract

Concentrated Solar Power (CSP) plants store energy by heating a storage medium with an array of mirrors that focus sunlight onto solar receivers atop a central tower. Operating at high temperatures these receivers face risks such as freezing, deformation, and corrosion, leading to operational failures, downtime, or costly equipment damage. We study the problem of anomaly detection (AD) in sequences of thermal images collected over a year from an operational CSP plant. These images are captured at irregular intervals ranging from one to five minutes throughout the day by infrared cameras mounted on solar receivers. Our goal is to develop a method to extract useful representations from high-dimensional thermal images for AD. It should be able to handle temporal features of the data, which include irregularity, temporal dependency between images and non-stationarity due to a strong daily seasonal pattern. The co-occurrence of low-temperature anomalies that resemble normal images from the start and the end of the operational cycle with high-temperature anomalies poses an additional challenge. We first evaluate state-of-the-art deep image-based AD methods, which have been shown to be effective in deriving meaningful image representations for the detection of anomalies. Then, we introduce a forecasting-based AD method that predicts future thermal images from past sequences and timestamps via a deep sequence model. This method effectively captures specific temporal data features and distinguishes between difficult-to-detect temperature-based anomalies. Our experiments demonstrate the effectiveness of our approach compared to multiple SOTA baselines across multiple evaluation metrics. We have also successfully deployed our solution on five months of unseen data, providing critical insights for the maintenance of the CSP plant. Our code is available at: https://tinyurl.com/ForecastAD

Figures (8)

• Figure 1: Visualisation of different properties of the data
• Figure 2: Examples of different types of normal and anomalous images
• Figure 3: Illustration of the end-to-end architecture of ForecastAD. The model is trained to forecast the next image in the sequence given a context embedding $c_i$ of $K$ prior data points obtained using a sequence-to-sequence model. For $(x_{i-k}, t_{i-k}, y_{i-k}) \in \mathcal{D}$ in the context, we sum the embeddings of inter-arrival time $\tau_{i-k}$ and interval since the start of operation $\delta_{i-k}$ and concatenate it with the image embedding. The anomaly score $s(x_{i}, t_{i})$ is computed as the difference between the forecasted and original image.
• Figure 4: Dataset split for two different training setups and the test set
• Figure 5: Ablation of different architectures