目前的异常检测模型存在训练不稳定、容易产生梯度消失的问题,影响异常检测效果,针对该问题,提出一种LSTM-WGAN模型,WGAN负责捕获变量之间的潜在关联,进一步提升了LSTM的检测能力。同时,以Wasserstein距离代替交叉熵损失训练判别器和生成器,结合重构损失以及判别损失实现异常检测。在NAB公开数据集上的实验结果表明LSTM-WGAN相较...
In this paper, we propose an anomaly detection model based on LSTM-WGAN. The multi-layer LSTM network captures temporal dependencies and introduces a self-attention layer to integrate embedded weather, holidays and other influencing factor data. Embed different features into the WGAN framework and ...
Speech Emotion Recognition on Small Sample Learning by Hybrid WGAN-LSTM NetworksSpeech emotion recognitionsmall sample trainingdata augmentationgenerative adversarial networklong short-term memoryThe speech emotion recognition based on the deep networks on small samples is often a very challenging problem in...
Long Short-Term Memory (LSTM), Variational Autoencoders (VAEs), and Wasserstein Generation Adversarial Networks (WGANs) were all used together in this study. This is a new way to make it easier to find problems in wind energy. The proposed method in this paper, based on the above VAEs...
This study compares deep learning methods for stock price prediction, including LSTM, GRU, GAN, and WGAN-GP. Using Google stock data, model performance was assessed via RMSE, MAE, and R metrics. The LSTM model achieved the best performance with the lowest error and highest...