Search

Anomaly detection (AD) plays a crucial role in identifying unusual data patterns indicative of potential issues or opportunities. Recent data-driven AD models require extensive training for satisfactory performance. This study explores the potential of untrained neural networks (UNNs) for AD tasks. UNNs are used for nonlinear random projection. The anomaly scores are derived from the randomly mapped features using the Mahalanobis distance. We conducted a series of experiments on 12 tabular and two image datasets, comparing the performance of UNNs with 12 established AD models, including state-of-the-art deep learning approaches. Our results demonstrate that UNNs can achieve competitive AD performance without training, which also underscores the importance of training to ensure higher performance beyond the untrained baseline. In addition, the proposed approach offers advantages in terms of time, computational costs, and accessibility, making it a compelling alternative for various applications.

As the power grid becomes more complex and dynamic, accurate short-term load forecasting (STLF) with probabilistic information is a prerequisite for various smart grid applications. For doing this, various deep learning models have been proposed, and recent models increase model size and complexity to achieve better accuracy which could also increase the burden on model design, computation time, and resources. To this end, we propose a novel deep learning model for accurate and efficient probabilistic STLF (PSTLF). First, we develop an STLF model utilizing the multi-layer perceptron (MLP)-mixer structure, i.e., MLP-mixer for STLF (MM-STLF), that has an advantage in forecasting accuracy and efficiency compared to the other deep learning models. Then, we propose a random quantile regression (RQR) method that takes a cumulative probability $\tau $ as an input to the model and is trained on random $\tau \text{s}$ . By combining MM-STLF and RQR, we develop a novel deep-PSTLF model, namely quantile-mixer (Q-mixer). We evaluate the overall performance of the proposed model with seven load datasets in terms of prediction error, model size, and inference time, respectively. Through experiments, various STLF models and probabilistic forecasting methods are compared, and the experimental results demonstrate the effectiveness of Q-mixer in load forecasting.