Aiming at the problem that the measurement accuracy of HMP45D temperature and humidity sensors used in automatic weather station is susceptible to temperature, the fitness function, selection, crossover, mutation in genetic algorithm (GA) is improved, the improved GA is used to optimize the penalty function, radial basis function and insensitive loss function in support vector machine (SVM). Based on the multiple sets of experimental data under different temperature and humidity, this method is used to establish a model and the results are compared with the traditional SVM regression model for temperature compensation. The experimental results show that the absolute error using the GASVM model is 0.1367%, reduced by 2.8351% than traditional SVM model. The proposed algorithm overcomes traditional SVM compensation model with low precision, slow process speed and has global optimization, convergence speed, higher compensation accuracy, effectively compensates temperature effect and greatly increases the measurement accuracy.