Helios-Artemis: Design and Simulation-Based Validation of a Dual- Microcontroller Predictive Solar MPPT Controller with On-Device LSTM Retraining for Sylhet Monsoon SHS Deployment in Bangladesh

Revised version of manuscript 25195, under review at the InternationalJournal of Power Electronics and Drive Systems (IJPEDS). **Abstract** This paper addresses the monsoon yield gap of the IDCOL Solar Home System(SHS) programme: conventional reactive Perturb-and-Observe (P&O) MPPTcontrollers incur persistent transient losses under the rapid cloudtransitions of the Sylhet monsoon. The Helios-Artemis dual-microcontrollerarchitecture partitions the problem — an ESP32-S3 (Helios) runs on-deviceLSTM irradiance forecasting with an adaptive gain scheduler, while anSTM32F103 (Artemis) executes real-time variable-step P&O over a 50 kHz buckconverter. Monte Carlo simulation over 30 stochastic July days yields a meanmonsoon tracking efficiency of 94.0% (+23.3 pp vs plain P&O), and 42 h offield irradiance logging from Sylhet validates the synthetic Markov+OU model'sshort-timescale variability regime. Estimated controller cost: ~1,750 BDT(USD 16), an 87% reduction versus commercial IDCOL-compatible controllers. **Links** - GitHub repository: https://github.com/touhidsiddiqueeraj-bit/artemis-helios- Journal: International Journal of Power Electronics and Drive Systems (IJPEDS), paper ID 25195 No external funding was received. This preprint is posted for publicdissemination of the revision; the manuscript is under consideration onlyat IJPEDS.