Deep Learning-Based MPPT for PV Systems: LSTM Forecasting and Adaptive TSMC via PPO Agent

Aymen Lachheb; Chabakata Mahamat; Rym Marouani

doi:10.2478/pead-2026-0018

.blurhash-client-img { display: none !important; }

Deep Learning-Based MPPT for PV Systems: LSTM Forecasting and Adaptive TSMC via PPO Agent

Power Electronics and Drives

Volume 11 (2026): Issue 1 (January 2026)

By: Aymen Lachheb , Chabakata Mahamat and Rym Marouani

Open Access

|Jun 2026

Figures & Tables

Overall architecture of the proposed LSTM–PPO–TSMC-based MPPT control scheme for a PV system. IGBT, Insulated Gate Bipolar Transistor; LSTM, long short-term memory; MOSFET, Metal-Oxide-Semiconductor Field-Effect Transistor; MPP, Maximum Power Point; MPPT, maximum power point tracking; PPO, proximal policy optimisation; PV, photovoltaic; PWM, pulse width modulation.

Flowchart of the proposed hybrid LSTM–PPO–TSMC MPPT. LSTM, long short-term memory; MPPT, maximum power point tracking; PPO, proximal policy optimization.

Architecture of the LSTM model for the prediction of maximum power voltage VMPP. LSTM, long short-term memory; VMPP, maximum power point voltage prediction.

PPO Agent training performance analysis. PPO, proximal policy optimisation.

Power comparison under variable irradiation. ANN, artificial neural networks; LSTM, long short-term memory; P&O, perturb & observe; PPO, proximal policy optimisation; PSO, particle swarm optimisation.

Power comparison of the MPPT algorithms under fluctuating irradiance. LSTM, long short-term memory; MPPT, maximum power point tracking; PPO, proximal policy optimisation.

Dynamic adjustment of TSMC parameters by the PPO Agent. PPO, proximal policy optimisation.

Frequency-domain analysis of control signal chattering. LSTM, long short-term memory; PPO, proximal policy optimisation.

Efficiency deviation vs. irradiance change. LSTM, long short-term memory; PPO, proximal policy optmisation.

Comprehensive computational complexity comparison of MPPT methods

MPPT method	Operations/step	Execution Time (ms)	Memory (KB)	Update rate	Relative cost
Classic SMC	32	0.02	2	Full switching	0.08×
TSMC	64	0.05	4	Full switching	0.13×
ANN–MPPT (3-layer)	1,280	0.18	12	1 kHz	0.70×
PSO-MPPT (10 particles)	3,200	0.35	28	100 Hz	3.10×
Full DRL (online)	18,000	2.10	256	10 Hz	11.3×
LSTM–PPO–TSMC (proposed)	4,416	0.45	38	1 kHz	1.0×

Performance metric for different MPPT algorithms

MPPT method	Settling time (ms)	Overshoot (%)
Classic SMC	160	9.5
LSTM + TSMC	85	3.8
LSTM + PPO + TSMC (proposed)	55	1.6

Parameters used for comparative MPPT methods

Method	Main parameters
SMC	λ = 5, switching gain = 10
TSMC	α = 0.8, σ = 0.5
ANN-MPPT	1 hidden layer, 10 neurons, tanh
PSO-MPPT	Particles = 10, w = 0.7, c₁ = c₂ = 1.5
P&O	Step size = 0.01
Proposed	LSTM (2 × 64), PPO adaptive α,σ

References

Authors

Metrics

Articles in this issue

DOI: https://doi.org/10.2478/pead-2026-0018 | Journal eISSN: 2543-4292 | Journal ISSN: 2451-0262

Journal RSS Feed

Language: English

Page range: 281 - 298

Submitted on: Jan 28, 2026

Accepted on: May 12, 2026

Published on: Jun 19, 2026

Published by: Wroclaw University of Science and Technology

In partnership with: Paradigm Publishing Services

Keywords:

deep reinforcement learning,

LSTM,

terminal sliding mode control,

optimisation algorithm,

PPO agent,

MPPT

Related subjects:

Computer sciences,

Artificial intelligence,

Engineering,

Electrical engineering,

Electronics

© 2026 Aymen Lachheb, Chabakata Mahamat, Rym Marouani, published by Wroclaw University of Science and Technology
This work is licensed under the Creative Commons Attribution 4.0 License.

Volume 11 (2026): Issue 1 (January 2026)

Deep Learning-Based MPPT for PV Systems: LSTM Forecasting and Adaptive TSMC via PPO Agent

Figures & Tables

Figure 1.

Figure 2.

Figure 3.

Figure 4.

Figure 5.

Figure 6.

Figure 7.

Figure 8.

Figure 9.

Comprehensive computational complexity comparison of MPPT methods

Performance metric for different MPPT algorithms

Parameters used for comparative MPPT methods

Paradigm

My account