Home // CSRF 2024, The First International Conference on Sustainable and Regenerative Farming // View article

Forecasting Agricultural Time Series Sensor Data Using Long-Short Term Memory Autoencoders

Authors:
Derick Wagambula

Keywords: time series forecasting; LSTM autoencoders; precision farming; wireless sensors.

Abstract:
Conventional Agriculture is evolutionizing towards Regenerative Farming, which involves a range of techniques supported by innovative technologies to address climate change. Among them is the IoT (Internet of Things) technology in agriculture, which has seen continuous streams of data in real-time. From the use of drones to deployment of Wireless Sensors in the field, data is collected and transmitted via a communication channel to an Internet of Things platform. In this paper, we analyze the use of digital tools in regenerative farming, specifically soil sensors, and demonstrate this with the use of Long-Short Term Memory (LSTM) autoencoders to forecast future sensor readings based on historical data which can help a farmer make better farming decisions. LSTM networks are a type of Recurrent Neural Networks (RNNs) and have the ability to capture long-term dependencies, handle complex patterns in sequential data, and learn from past errors. This is evident through their use in predicting household power consumption, network traffic speed prediction, and predicting the crop yields. The proposed model is applied to the Cook Agronomy Farm (CAF) dataset, which contains field-scale sensor dataset for soil moisture and soil temperature at various levels. Using the Root Mean Square Error (RMSE) to evaluate the performance, the proposed model takes in multiple features as input and forecasts multiple steps and multiple parallel features. Traditional models such as Autoregressive Integrated Moving Average (ARIMA) have been used to forecast multivariate time series data. However, the proposed LSTM autoencoders perform with high accuracy and robustness in forecasting agricultural sensor data.

Pages: 16 to 22

Copyright: Copyright (c) IARIA, 2024

Publication date: November 17, 2024

Published in: conference

ISBN: 978-1-68558-323-1

Location: Valencia, Spain

Dates: from November 17, 2024 to November 21, 2024