Implementasi Prediksi Suhu menggunakan Akselerasi Systolic Array dengan Machine Learning pada Edge Computing

Abstract

As the volume of data from Internet of Things (IoT) devices increases, efficient data processing at the edge has become a must to overcome latency and bandwidth limitations. This study aims to design and implement an accelerated Artificial Neural Network (ANN)-based temperature prediction system using systolic array architecture on the Kria KV260 Field-Programmable Gate Array (FPGA) platform for edge computing applications. The ANN model used is the Feedforward Multilayer Perceptron which is trained using historical temperature data. The 9x9 systolic array architecture is designed to accelerate the matrix multiplication operations that are dominant in ANN inference. The system implementation is carried out by integrating the hardware accelerator on the FPGA with the Zynq UltraScale+ MPSoC processor system. The results of the CPU test showed that the ANN model achieved high accuracy with a Mean Squared Error (MSE) of 0.0313, Mean Absolute Error (MAE) of 0.1417, and an R-squared (R²) of 0.9457. Hardware evaluations show efficient resource utilization and low power consumption, proving that acceleration using systolic arrays on FPGAs is an effective and high-performance solution for real-time machine learning applications in edge computing environments