In this study, an improved northern goshawk optimization algorithm called SR-NGO is developed to solve the parameter identification problem of the Single Diode Model (SDM), Double Diode Model (DDM), and Triple Diode Model (TDM). Compared to the NGO algorithm, the SR-NGO algorithm reduces the minimum, maximum, mean, and standard deviation values of the root mean square error between experimental I-V data and the SDM(DDM, TDM) fitting data by 0.0004284% (0.2542%, 0.2034%), 1.1094% (3.7313%, 6.0740%), 0.05106% (0.6901%, 1.0699%), and 99.9999% (88.1540%, 94.0600%), respectively. Furthermore, the SR-NGO algorithm exhibits higher accuracy, faster convergence speed, and stronger stability compared to other existing algorithms. These results demonstrate that the SR-NGO algorithm, which combines the sine cosine guiding mechanism and random learning mechanism, enhances both local and global search capabilities, thereby improving the ability to escape local optima.