HVAC Indoor Temperature Forecasting

Table of Contents

• Introduction
• Supported Combinations
• Dataset and Model Details
  • Dataset
  • Model Architecture
  • Input Features
  • Output
• Feature Extraction Configuration
• Steps to Run the Example
• See Also
• Sample Output

Introduction

Traditional HVAC controllers are reactive - they respond to temperature deviations only after they occur, leading to delayed corrections, unnecessary compressor cycling, and wasted energy. A smarter approach is to anticipate what the indoor temperature will be in the near future and act preemptively.

This project demonstrates on-device machine learning for HVAC control. A compact LSTM model running directly on the AM26x MCU predicts upcoming temperature trends from recent sensor readings, enabling smoother, more energy-efficient control decisions without relying on cloud connectivity.

Supported Combinations

Parameter	Value
CPU + OS	r5fss0-0 nortos
Toolchain	ti-arm-clang
Board	am261x-lp
Example folder	examples/ai/hvac_indoor_temp_forecast/

Dataset and Model Details

Dataset

This example uses a synthetic HVAC dataset designed to emulate realistic indoor thermal dynamics. The dataset captures the interplay between compressor operation, outdoor conditions, and the resulting indoor temperature.

Parameter	Value
Signals	compressorFrequency, outdoorTemperature, indoorTemperature
Input Variables	3
Prediction Target	indoorTemperature (next timestep)

The model learns to predict the indoor temperature at the next timestep given the past 5 readings of all three signals.

Model Architecture

The model is FCST_LSTM10, a single-layer LSTM network with a hidden size of 10 and approximately 611 trainable parameters. LSTM gating mechanisms capture the temporal dependencies between compressor activity, outdoor conditions, and the resulting indoor temperature trajectory.

Input Features

The model takes 4D input (N,C,H,W):

• N (1): batch size, restricted to 1
• C (3): channels - compressorFrequency, outdoorTemperature, indoorTemperature
• H (5): number of historical timesteps (frame size) used for prediction
• W (1): width, restricted to 1 for timeseries applications

Output

The model produces an output tensor of shape (1, 1, 1, 1), representing the predicted indoor temperature at the next timestep (forecast_horizon = 1). This is a continuous forecasting output that can be fed directly into a downstream control algorithm.

Feature Extraction Configuration

A SimpleWindow transform segments the continuous stream into fixed-length windows of 5 consecutive timesteps across all 3 input channels. No FFT, binning, or log scaling is applied - the LSTM model learns directly from raw sensor values.

Steps to Run the Example

• When using CCS projects to build, import the CCS project for the required combination and build it using the CCS project menu (see Using SDK with CCS Projects).
• When using makefiles to build, note the required combination and build using make command (see Using SDK with Makefiles)
• Launch a CCS debug session and run the executable, see CCS Launch, Load and Run
• The application will forecast indoor temperature and compare against expected values

See Also

AI Examples

Sample Output

HVAC Indoor Temperature Forecasting Example Started ...
All tests have passed!!
Golden vectors matched: 3 not matched: 0