Skip to main content

Model Introduction

Time Series Forecasting Research​

In recent years, multiple studies have explored the application of Large Language Models (LLMs) in time series forecasting, laying a solid academic foundation:

  • Time-LLM: Time Series Forecasting by Reprogramming Large Language Models, ICLR 2024 discusses how to use reprogramming of large language models for time series forecasting, demonstrating the potential of LLMs in capturing complex temporal dependencies.

  • Time-MMD: A New Multi-Domain Multimodal Dataset for Time Series Analysis, NeurIPS 2024 introduces a new multi-domain multimodal dataset that provides rich resources for time series analysis, advancing research in various time series prediction tasks.

  • Large Language Models Are Zero-Shot Time Series Forecasters, NeurIPS 2023 shows that LLMs can perform time series forecasting without specific training, showcasing the effectiveness of their zero-shot learning capability.

These studies not only provide a theoretical basis for the application of LLMs in time series forecasting but also demonstrate the cutting-edge advancements in the field.

Algorithm Model​

The XMO-Time 1.0 base model is specifically designed for photovoltaic power prediction in the electricity market. This model has been pre-trained on a large amount of photovoltaic data, demonstrating excellent performance. It supports predicting photovoltaic power from 1 day to 3 years, enabling you to quickly and efficiently grasp the future generation trends and output of photovoltaic plants.

In the rapidly changing electricity market, accurate time series forecasting is crucial for decision-making. XMO focuses on developing advanced time series prediction models that integrate the latest machine learning and deep learning architectures, especially the application of large models. Our model can handle large-scale data from multiple sources and effectively capture complex temporal patterns, providing high-precision short-term and long-term forecasts.

Our algorithm is adaptive, capable of real-time updates based on market changes and new data to ensure the accuracy and reliability of forecasting results. By using large models, we can analyze and process multidimensional feature data, significantly improving prediction performance. Our solution is widely used in electricity demand forecasting, renewable energy generation prediction, and related fields, providing clients with robust decision support to help them optimize operations and resource allocation.