Researchers have published a coordinated control framework for solar-powered electric-vehicle charging stations that combines adaptive maximum power point tracking, Kalman-filter battery state estimation, predictive energy dispatch and a stability analysis of the station's switching converters. The study, reported in Nature's Scientific Reports, comes from authors affiliated with SRM TRP Engineering College and Sri Eshwar College of Engineering in Tamil Nadu, Hindustan Institute of Technology in Coimbatore, and Kampala International University in Uganda.
The paper describes a station architecture with photovoltaic panels, battery storage and a grid connection, and reports results across four technical layers. According to Nature, only two of those layers — the tracking algorithm and the state-of-charge estimator — were built and tested on hardware.
Why it matters
Solar-assisted charging stations must juggle three things at once: harvesting as much panel output as conditions allow, knowing how much charge the battery actually holds, and deciding when to draw from the grid versus the battery. Each of these is a separate engineering problem, and the paper's argument is that treating them together, with formal guarantees on the power electronics, produces a more dependable system than tuning each piece in isolation.
The reported efficiency figures are simulation-level for the full system. Even so, the authors position the work as a step toward charging infrastructure that leans less on the grid — a persistent concern for operators in regions where supply is unreliable or expensive.
Adaptive tracking beats fixed-step methods in randomised tests
The first contribution is a perturb-and-observe maximum power point tracking algorithm whose step size is adjusted by a fuzzy logic layer. Instead of nudging the operating voltage by a fixed increment, the controller scales its adjustment according to how sensitive the local power–voltage curve is, which the authors say lets a single parameter set handle changing sunlight.
Across 15 randomised irradiance scenarios, the fuzzy-adaptive tracker reached 98.1% tracking efficiency with a margin of 0.1 percentage point, according to the paper. Fixed-step perturb-and-observe came in at 95.0% with a wider spread of 0.6 points, while a linear-adaptive method re-tuned separately for each scenario reached 97.8%. Nature reports the comparison against the re-tuned linear law was statistically significant, with a Welch's t-test value of 5.2 and p below 0.0001.
The practical point is the single parameter set: the linear-adaptive baseline needed retuning per scenario to stay close, whereas the fuzzy version did not.
Battery state estimation and predictive dispatch
The second layer is an extended Kalman filter that estimates battery state of charge by combining Coulomb counting with a second-order Thevenin equivalent circuit model of the battery. The authors report that a Popov–Belevitch–Hautus test confirms the model pair is observable, meaning the state can in principle be recovered from the measurements available.
Over 20 Monte Carlo trials, the filter cut mean root-mean-square state-of-charge error from 2.45% to 0.118%, per the paper.
The third layer swaps static rule-based dispatch for model predictive control, which re-optimises over a moving horizon. Under an assumption of perfect forecasting, the paper reports a 6.8% reduction in operating cost. The authors also note that this gain comes with an explicit trade-off against grid independence — the two objectives pull against each other rather than improving together.
The fourth contribution is a stability certificate. By constructing Lyapunov functions that depend on the converter's operating mode, the authors derive an average dwell-time bound of 2.31 ms. Any switching sequence that respects that bound keeps the linearised converter dynamics exponentially stable, according to the analysis.
Hardware gap leaves predictive control unverified
The full simulation reports 93.6% system efficiency, DC-bus voltage held within plus or minus 5%, and grid dependency of 14.2%.
Only the fuzzy-adaptive tracking and the Kalman-filter state estimation were implemented on an ESP8266 prototype. On that subset, measured system efficiency was 91.2%, which the authors say agrees with the simulated 93.6% to within 2.4 percentage points.
The predictive dispatch results and the Lyapunov stability proof remain simulation-level findings. The paper names hardware validation of both as a priority for future work — a caveat worth keeping in view, since the cost-saving figure rests on an idealised forecast and the stability bound has not yet been exercised on real converters.
The article is published open access under a Creative Commons Attribution-NonCommercial-NoDerivatives licence, and Springer Nature states it remains neutral on jurisdictional claims in the published maps and institutional affiliations.
Watch for follow-up work from the group on hardware testing of the predictive controller and the dwell-time stability bound.
