In this paper, inspired by the T cell immune process, a new meta-heuristic algorithm named T cell immune algorithm (TCIA) is proposed to better solve various complex practical problems. The core difficulty of meta-heuristic algorithms is balancing exploration and exploitation. TCIA mimics the behavior of T-cell immune process in recognizing antigens, activating cells and attacking pathogens, thus to better balance exploration and exploitation. Recognition is the process of judging the cell concentration at the current location, which is used to evaluate the fitness value of the current cell. Activation is the process by which T cells are activated and thus differentiate into multiple cells, helping the population to explore on a larger scope. Attacks are divided into random attack and directed attack. Random attack helps the population escape from local optima by randomly selecting a target cell to attack. Targeted attacks push the overall approach to the location with the highest concentration of target cells, which makes better use of the global optimal solution. The TCIA is tested on CEC2022 and compared with other 10 algorithms. Experimental results and statistical analysis show that TCIA is more effective than other 10 classical and new meta-heuristic algorithms in solving constraint function problems. TCIA is tested on three multi-objective functions, which verifies its good ability in solving multi-objective functions. Two classical engineering design problems are applied to verify the ability of TCIA to solve practical problems. The parameters of Back propagation neural network are optimized by TCIA to predict the compressive strength of concrete. The results show that the ability of TCIA to solve a variety of problems is balanced and excellent.

Parameter identification of photovoltaic models (PIPM) is essential for controlling a photovoltaic (PV) system. However, due to its complexity, most existing methods still suffer from problems such as low accuracy, sensitivity to initial values and local optima. For this, an improved butterfly optimization algorithm (DLBOA) with dimension differential learning is proposed. First, a new adaptive fragrance is introduced to optimize the instability caused by target differences and improve convergence performance. Second, the proposed dimension differential learning strategy improves butterflies’ position by utilizing the excellent dimension information within population, thereby reinforcing inter-individual learning and enhancing population balancing and diversity, ultimately escaping from local optima. Then, after evaluating based on CEC2022, DLBOA identified the parameters for 8 models across 5 materials, and outperformed 9 state-of-the-art algorithms in terms of accuracy, robustness, and promoting percentage. DLBOA is further compared with 9 existing PIPM methods including 5 numerical methods. Finally, applying DLBOA to the YL PV station in China Guizhou Power Grid under dynamic climate, multiple metrics confirm DLBOA’s outstanding accuracy, with the reconstructed I-V and P-V curves closely matching synthesized curves. The statistical analysis results demonstrate that the proposed method effectively enhances the robustness of parameter identification while also strengthening the ability to escape local optima, demonstrating the potential to improve PIPM accuracy.