Multi-Agent Time Series Network

I don't produce prompts, I'm just a prompt transporter.

When debugging wind power prediction models, we need to try various parameter and feature combinations, which is time-consuming and laborious. We use multi-agent collaboration combined with LLM model optimization capabilities to automate model training and tuning.

Planning 5 Agents: Data Analysis Agent, Feature Selection Agent, Model Tuning Agent, Training Evaluation Agent, Result Analysis Agent. They trigger each other's execution through OpenAgents channel messaging. Planning to use OpenAgents message listening events for agent interaction coordination, allowing users to track actual training progress in conversations. Also using OpenAgents LLM calling capability for Agent work.

1. Agent A - Data Analysis Agent: Loads and preliminarily analyzes time series data (user uploads CSV files), extracts basic info like shape, time range, column correlations. 2. Agent B - Feature Selection Agent: Selects appropriate features based on feature library and LLM suggestions (LLM generates feature engineering code). 3. Agent C - Model Tuning Agent: Uses LLM to recommend optimal hyperparameters for LightGBM and XGBoost (LLM generates network structure and parameters). 4. Agent D - Training Evaluation Agent: Executes actual model training and evaluation (combines Agent B and C results to train and evaluate code). 5. Agent E - Result Analysis Agent: Analyzes each iteration's results, plots performance trend charts, decides whether to continue next iteration.