Can grey prediction models effectively forecast geothermal heat pump temperatures using very small datasets?

Conditions

• The dataset consists of a small number of sequential measurements (e.g., 24 data points).
• The goal is short-term forecasting (e.g., predicting 8 hours ahead).

Reasoning, step by step

1. Collect a small dataset of 24 hourly temperature measurements from a specific GHPS location.
2. Train the IGM (1,1) model on this 24-hour dataset.
3. Use the trained model to forecast the output temperature for the next 8 hours.
4. Compare the predicted values against actual measurements or baseline GM (1,1) predictions using RSME, MAE, and MAPE.

Example

It was deployed to predict output temperatures using real-world data from three distinct institutions: Oklahoma University, Universitat Politècnica de València, and Oakland University. For each location, the model utilized a remarkably small dataset consisting of just 24 hourly data points to train and subsequently predict the temperature eight hours into the future.

Common misconceptions

• Believing that machine learning or forecasting models always require massive datasets to be effective.
• Assuming that grey prediction models cannot handle real-world noise or variability in small samples.

Watch the explanation

Explore next

• How does the IGM (1,1) model correct its current predicted value compared to the standard GM (1,1) model?
• What statistical metrics are used to evaluate the accuracy of time series forecasting models like IGM (1,1)?
• What are the anticipated future improvements for the IGM (1,1) model regarding geothermal heat pump systems?