Lyrebird-inspired AI controller slashes solar inverter distortion to 0.29%
Engineers have unveiled a solar power conversion system whose control software is tuned by an algorithm modelled on the defensive antics of the Australian lyrebird, and the results are striking. In a study published in Discover Energy, Satyaprasad Mohapatra, Akshaya Kumar Patra and Sanjeeb Kumar Kar report that their hybrid controller cut the total harmonic distortion of an inverter’s output current from 8.59 percent in open-loop operation to just 0.29 percent once the closed-loop scheme was engaged. The work, published on 2 November 2026, combines a photovoltaic array, an interleaved super-lift Luo DC-DC converter and a transformerless H6 inverter into a single standalone solar photovoltaic energy conversion architecture, all orchestrated by an unusually layered control strategy.
The centrepiece of the design is a controller the authors call LOA-H-ANFIS-2DoF-FOPIDN, a name that packs together several control-engineering traditions. It merges the adaptive learning capability of an Adaptive Neuro-Fuzzy Inference System, or ANFIS, with the design flexibility of a two-degree-of-freedom fractional-order PID controller equipped with a noise-attenuating filter, abbreviated 2DoF-FOPIDN. Every parameter of this hybrid controller is optimised offline by the Lyrebird Optimizer Algorithm, a metaheuristic inspired by the way lyrebirds choose between escaping danger and hiding from it. On the alternating-current side, the H6 inverter is governed by a second LOA-optimised PI controller working in tandem with a multi-carrier sinusoidal pulse-width modulation scheme, which the team credits for the stable output voltage and reduced harmonic content.
The Lyrebird Optimizer Algorithm itself is a relative newcomer among nature-inspired optimisers. Each candidate solution is treated as a lyrebird position vector, and at every iteration the algorithm probabilistically selects between two update mechanisms. The escape strategy drives exploration, pushing members toward regions of the search space occupied by solutions with better objective values, while the hiding strategy handles exploitation through small, deliberate movements that refine positions near protected spots. The authors configured the optimiser with 50 particles, 15,000 iterations and 30 replications, and set it to minimise the integral of time-weighted absolute error, a performance index that penalises both transient deviations and long-term tracking inaccuracy. The recorded computation time for the optimiser was 75.4034 seconds under their MATLAB configuration, with eight decision variables: six belonging to the 2DoF-FOPIDN controller and two to the inverter-side PI controller.
The ANFIS portion of the hybrid controller follows a Takagi-Sugeno inference structure organised into five computational layers, from input mapping through membership evaluation, rule activation, normalisation and output synthesis. Gaussian membership functions were chosen for their smooth, differentiable character, and each of the two ANFIS models, one generating the alpha parameter and one the beta parameter, implements seven membership functions per input variable, yielding 49 fuzzy rules through grid partitioning. Training used 500 samples split 70:15:15 into training, validation and testing subsets, with a hybrid learning algorithm combining gradient-based backpropagation and least-squares estimation run over 150 epochs. The resulting adaptive surfaces enforce stronger control action during transients and gentler corrections near steady state, damping oscillations as the converter settles.
Why go to such lengths on the DC side? The interleaved super-lift Luo converter is tasked with stepping the photovoltaic array’s output up from 100 volts to 400 volts, and its cascaded energy-pump mechanism with phase-shifted interleaving achieves high voltage gain at low duty ratios while keeping ripple low. Under open-loop conditions, driven only by a 10 kHz pulse generator with no feedback, the converter exhibited a ripple voltage of 30 volts, a shortcoming the authors attribute to the nonlinear characteristics of the solar module and the absence of any controlling action. Closed-loop operation with the proposed controller changed the picture dramatically: the regulated 400-volt output settled in 0.0022 seconds, rose in 0.0023 seconds, overshot by 12.22 percent and left a steady-state error of just 0.00011 volts, figures that matched closely between the MATLAB/Simulink simulation and the real-time hardware validation.
That real-time validation is a notable strength of the study. Rather than stopping at simulation, the team executed the complete solar photovoltaic system and its control algorithms on an OPAL-RT 4510 real-time simulator integrated with a Xilinx Kintex-7 FPGA, providing hardware-in-the-loop evidence of real-time feasibility. The platform offers deterministic execution and synchronisation between control and measurement channels, which the authors argue makes the evaluation a closer reflection of practical operating conditions than simulation alone. Photographs of the experimental arrangement accompany the paper, and the hardware-in-loop results for both the converter and the inverter tracked the simulated responses closely.
The inverter results are where the harmonic story becomes concrete. Supplying a series R-L load with resistance of 34.74 ohms and inductance of 87.85 millihenries, corresponding to roughly 1.014 kilowatts of active power at 50 hertz, the closed-loop H6 inverter delivered a stable peak output voltage of approximately 339 volts and a load current of 7.64 amperes. The corresponding RMS values were 239.7 volts and 5.40 amperes, with voltage and current lagging by about 38.5 degrees to give a power factor of 0.783 and an apparent power near 1.294 kilovolt-amperes. Total harmonic distortion measured 0.29 percent, evaluated over 30 fundamental-frequency cycles using FFT analysis up to 1000 hertz, the 20th harmonic of the 50-hertz fundamental. The same figures emerged from the OPAL-RT hardware-in-loop runs, reinforcing the simulation findings.
Robustness testing pushed the system well beyond its nominal operating point. The team varied temperature from 18 degrees Celsius to 48 degrees Celsius and irradiance from 850 watts per square metre to 1350 watts per square metre, tabulating the resulting RMS output voltage, current and power, which remained nearly constant thanks to the closed-loop regulation. Step load changes were imposed by switching between identical series R-L loads connected in parallel, and the controller restored the terminal voltage to approximately 339 volts after each transition, with current amplitudes of 7.64 amperes for a single load and 15.28 amperes when both loads were connected. Under nominal conditions the photovoltaic input power was approximately 2.102 kilowatts against total output power of approximately 2.028 kilowatts, an efficiency of about 96.47 percent. The controller also held stable responses under plus-or-minus 20 percent plant parameter variations, converging to the desired operating point with only marginal changes in overshoot and settling time.
Stability analysis added a frequency-domain perspective. Bode plots of the closed-loop system showed a gain margin of 8.56 decibels and a phase margin of 62.36 degrees, which the authors interpret as adequate stability reserves against system dynamics, while a closed-loop pole-zero map confirmed the absence of poles in the right-half plane. Comparative step-response analysis against conventional controllers including ANN, fuzzy, hysteresis, FOPID and ANFIS schemes highlighted the proposed controller’s fastest settling behaviour at approximately 0.0022 seconds. The authors note that the 2DoF framework decouples reference tracking from disturbance rejection, that the fractional-order integrator and differentiator extend tuning flexibility across the frequency domain, and that the filtering stage attenuates high-frequency measurement noise, protecting the adaptive learning mechanism’s stability and convergence.
The study’s scope deserves a clear-eyed reading. This is a modelling, simulation and hardware-in-the-loop validation effort for a standalone solar photovoltaic system, not a field deployment on a rooftop or a utility-scale plant, and the authors themselves list full hardware prototyping, integration of maximum power point tracking with emerging intelligent control, and extension to three-phase systems as future work. The component modelling assumed ideal switching devices, identical interleaving branches and negligible semiconductor losses, and the photovoltaic cell model adopted standard simplifications including uniform irradiance distribution. The comparative assessment against fuzzy logic-PI, Type 2 fuzzy logic, ANN-PI, ANN-PID, ANFIS and ANFIS-FOPIDN schemes is presented as contextual evidence of effectiveness rather than a like-for-like benchmark under identical conditions. Even with those caveats, the headline numbers stand: harmonic distortion reduced by a factor of nearly thirty, a converter that settles in just over two milliseconds, and an optimiser that borrows its search strategy from a bird famous for outsmarting predators. For a renewable energy landscape demanding ever cleaner power electronics, that combination of intelligence, speed and measured rigor is exactly the kind of engineering story worth watching as it moves from the real-time simulator toward physical hardware.
Subject of Research: Hybrid intelligent control of a standalone solar photovoltaic converter system using a lyrebird-optimised ANFIS and fractional-order PID controller
Article Title: Advanced adaptive control of photovoltaic systems through LOA-tuned hybrid ANFIS–2DoF-FOPIDN
Article References: Mohapatra, S., Patra, A. K., & Kar, S. K. (2026). Advanced adaptive control of photovoltaic systems through LOA-tuned hybrid ANFIS–2DoF-FOPIDN. Discover Energy, 6(1), Article 98. https://doi.org/10.1007/s43937-026-00215-7
Image Credits: AI Generated
DOI: 10.1007/s43937-026-00215-7
Keywords: solar photovoltaic, ANFIS, fractional-order PID, Lyrebird Optimization Algorithm, H6 inverter, total harmonic distortion, DC-DC converter, hardware-in-the-loop, OPAL-RT, power electronics, voltage regulation, renewable energy
News Source: Denise Maddox. (November 7, 2026). Lyrebird-inspired AI controller slashes solar inverter distortion to 0.29%. Scienmag.
We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.
Enter your email address to subscribe to this blog and receive notifications of new posts by email.
Email Address
Subscribe
Bioengineer.org © Copyright 2023 All Rights Reserved.
Login to your account below


Remember Me

Please enter your username or email address to reset your password.


Bioengineer.org © Copyright 2023 All Rights Reserved.

source