Deep Learning-Based MPPT Approach to Enhance CubeSat Power Generation

The Electrical Power System (EPS) is a vital subsystem in CubeSats, responsible for powering all onboard components. Due to standardized size and weight constraints, the limited surface area of solar panels restricts power generation. To address this, Maximum Power Point Tracking (MPPT) is crucial. However, traditional MPPT techniques struggle in CubeSats’ dynamic orbital environments, where solar irradiance varies across different facets. This paper presents a novel MPPT method that combines a Deep Feedforward Neural Network (DFFNN) with a Proportional-Integral (PI) controller to adapt to these rapidly changing conditions. A year-long simulation of a 3U CubeSat in Systems Tool Kit (STK) generated real-time orbital temperature(T) and irradiance(G) data, producing 70,000 data points. The dataset was divided into 70% for training, 15% for testing, and 15% for validation to develop the DFFNN model. The trained DFFNN was integrated into a Simulink model and tested under variable illumination conditions. The results showed that the DFFNN-based MPPT achieved a 97.4% efficiency, outperforming Perturb and Observe (P&O) at 88.9%, Incremental Conductance (InC) at 89.7%, and Particle Swarm Optimization (PSO) at 94.08%. Additionally, the proposed method achieved a system efficiency of 88.3%, reduced power ripple to less than 2.5%, and significantly improved transient performance. These findings confirm the effectiveness of the DFFNN-PI approach in optimizing power extraction for CubeSat missions under dynamic space conditions.