Accurately grasping the total nitrogen content of farmland soil is significant for evaluating soil fertility and applying nitrogen fertilizer reasonably. To comprehensively utilize the advantages of each single prediction Model, improve the overall prediction performance, reduce the variance of the model, and improve the robustness, this study takes farmland brown soil as the research object, and based on near-infrared and visible hyperspectral data, puts forward a Combined prediction model based on standard deviation. CPM was used to predict soil total nitrogen content. Savitzky-Golay smoothing and first-order differential transformation are applied to the original hyperspectral data, and a tree model is used for feature band extraction. Using five single prediction models, Decision Tree Regression (DTR) (Model 1), Gaussian Kernel Regression (GKR) (Model 2), Random Forest Regression (RF) (Model 3), LASSO Regression (Model 4), and Multi-Layer Perceptron (MLP) (Model 5), a combination prediction model is established through a linear combination of single prediction models. The results indicate that: (1) The weights of the five single prediction models in the combined prediction model are obtained by generalized reduced gradient optimization algorithm: ω*1=0.407，ω*2=0.378，ω*3=0.215，ω*4=0，ω*5=0; (2) For all data, the predictive effectiveness of five single prediction models and combined prediction models for predicting soil total nitrogen content is M, respectively M1=0.855，M2=0.856，M3=0.847，M4=0.785，M5=0.796，MCPM=0.880, compared to the maximum predictive validity of a single model, the predictive validity of the combination prediction model has increased by 2.924%; (3) For all data, the prediction accuracy and standard deviation of soil total nitrogen content based on five single prediction models and combined prediction models are E(A1)=0.924，σ(A1)=0.075，E(A2)=0.928，σ(A2)=0.077，E(A3)=0.923，σ(A3)=0.082，E(A4)=0.882，σ(A4)=0.109，E(A5)=0.889，σ(A5)=0.104，E(ACPM)=0.937，σ(ACPM)=0.066, compared to the maximum prediction accuracy of a single model, the combination prediction model has improved prediction accuracy by 0.970% and model stability by 12.000%, making it an optimal combination prediction model. The combined prediction model can effectively estimate the total nitrogen content of farmland brown soil based on visible-near-infrared spectral data and can provide a basis and reference for the rapid monitoring of the total nitrogen content of farmland soil.