Research on MPPT control strategy based on LSTM and IHOA algorithm

• A novel hybrid MPPT strategy, LSTM-IHOA, is proposed for photovoltaic systems under dynamic and partial shading conditions. • The LSTM network provides accurate duty cycle reference values based on real-time irradiance and temperature data. • An improved Hiking Optimization Algorithm (IHOA) integrates Cauchy mutation, boundary particle recovery, and leader-following mechanisms. • LSTM-IHOA reduces tracking time by approximately 30 % while maintaining tracking efficiency above 99.84 %. • Simulation results demonstrate that LSTM-IHOA outperforms PSO, HOA, AOA, and SCSO in tracking accuracy, convergence speed, and robustness. To obtain a more efficient photovoltaic system, it is inevitable to adopt optimum maximum power point tracking (MPPT) algorithm. Swarm intelligence optimization algorithm is one of the most commonly used MPPT algorithms. However, they are prone to getting trapped in local optima, with slow convergence speed. This paper proposes a novel hybrid MPPT strategy (LSTM-IHOA) that integrates a Long Short-Term Memory (LSTM) network with an Improved Hiking Optimization Algorithm (IHOA). First, the proposed MPPT utilizes the LSTM to predict real-time duty cycle references using environmental parameters (irradiance and temperature), guiding IHOA with accurate initial search directions. Then, IHOA is applied to accurately find the duty cycle corresponding to the maximum power point. Especially, IHOA employs a Cauchy mutation strategy to escape local optima, incorporates a boundary particle recovery mechanism based on the duty cycle reference to constrain search space, and adopts a leader-following mechanism to accelerate local convergence. Quantitative validations under six operating scenarios demonstrate its superiority: the proposed MPPT strategy (LSTM-IHOA) can reduce the tracking time (reduced by 30 %), with power tracking achieving higher accuracy, smaller power oscillations, and greater efficiency (no <99.96 % tracking efficiency), outperforming four compared algorithms in both static and dynamic scenarios.