The elemental composition and content in Martian soil are important record carrier of geological evolutionary history, which can reflect the Martian environment, climate, and other information, so it is of great significance to detect and analyze Martian soil. A LIBS quantitative analysis method based on the combination of transfer component analysis (TCA) with random forest (RF) is proposed to predict the K₂O mass fraction of Mars on-orbit standards. The spectral data of 383 standard samples in simulated Martian environment were selected as the training set, and the spectral data of 6 on-orbit standard samples in real Martian environment were selected as the test set. The RF model with 250 decision trees was established using the training set, and the mean absolute error (E_MA), the root mean square error (E_RMS) and the mean relative error (E_MR) were 1.117, 1.148 and 10.104, respectively, indicating poor prediction performance. To shorten the distribution distance between the spectral data of the training set and the test set, the TCA-RF model is established and the parameters are adjusted. Compared with the RF model, the E_MAE_RMS and E_MR of the TCA-RF model are reduced by 90.7%, 88.1% and 94.1% respectively. Compared with the reference model MOC, a model based on the partial least squares regression combined with independent component analysis, the TCA-RF model is more accurate than the MOC model in predicting samples with K₂O mass fraction higher than or equal to 0.15% in the test set. Therefore, it is indicated that the TCA-RF model can provide a new technical means for detecting the content of soil elements on Mars.