MQ-ReTCNN: Multi-horizon time series forecasting with retrieval-augmentation

Multi-horizon probabilistic time series forecasting has wide applicability to real-world tasks such as demand forecasting. Recent work in neural time-series forecasting mainly focus on the use of Seq2Seq architectures [25]. For example, MQTransformer [10] – an improvement of MQCNN [27] – has shown the state-of-the-art performance in probabilistic demand forecasting. In this paper, we consider several methods to enhance model performance by incorporating cross-entity information and propose adding a cross-entity attention mechanism along with a retrieval mechanism to select which entities to attend over. We demonstrate how our new model, MQ-ReTCNN, leverages the encoded contexts from a pretrained MQCNN model on the entire population to improve forecasting accuracy. Using MQCNN as our baseline model (due to computational constraints, we do not use MQTransformer), we first show on a small demand forecasting dataset that it is possible to achieve ∼3% improvement in test loss by adding a cross-entity attention mechanism where we attend over all other entities in the population. We then evaluate the model with our proposed retrieval mechanism – as a means of approximating an attention over a large population – on a large-scale demand forecasting application with over 2 million products and observe ∼1% performance gain over the MQCNN baseline.