我国西北地区具有煤层厚度大、埋藏浅等特点，无人机热红外监测与温度补偿是浅埋煤火灾害信息监测的关键技术，对推进煤自燃灾害安全监测与影响范围预警评价有着重要意义。针对复杂环境下多参数对热红外温度监测结果的影响，提出了一种基于灰狼优化—双层广义回归网络与连续高度修正函数相结合的温度补偿模型。基于热红外装置中大气消光系数对接收辐射对比度的影响，选取热红外监测波段下影响温度结果的多种因素，并通过主成分分析法确定累计能够表征实际温度的指标因子，以多种环境下的无人机热红外监测实验中指标因子数据为GWO-双层GRNN网络输入，得到训练完成的离散温度补偿模型，并提出无人机高度的连续修正函数作为模型输入待补偿数据的前置流程，最后将完整温度补偿模型进行了试验与现场验证。结果表明，GWO-双层GRNN网络在数据测试中，离散补偿效果优于其他模型，达到了平均绝对误差≤0.008 1、均方根误差≤0.013 2、决定系数≥0.996 9，表明该模型具有良好的补偿效果；连续修正函数避免了无人机高度对热红外监测结果的影响，将无人机连续高度热红外温度回归问题划分为阶跃式回归计算，且最终模型具有良好的监测精度，提高了温度补偿模型的泛化能力。为应用无人机热红外监测结果划分浅埋煤火危险区域提供了配套计算方法，可进一步将该方法推广至对应的无人机应用与激光监测行业。

The northwestern region in China has the characteristics of large coal seam thickness and shallow burial depth. Unmanned aerial vehicle (UAV) thermal infrared monitoring and temperature compensation are key technologies for monitoring shallow coal fire disasters, which is of great significance for advancing the safety monitoring and warning evaluation of coal spontaneous combustion disasters. A temperature compensation model based on the grey wolf optimization-double-layer generalized regression neural network (GWO-double-layer GRNN) and continuous altitude correction function is proposed to address the influence of multiple parameters on thermal infrared temperature monitoring results in complex environments. The factors affecting the temperature results under the thermal infrared monitoring band are selected based on the contrast of the atmospheric extinction coefficient to the received radiation, and the index factors that can cumulatively characterize the actual temperature are determined by principal component analysis. The index factor data from the UAV thermal infrared monitoring experiments in various environments are used as inputs to the GWO-double-layer GRNN network to obtain a discretely trained temperature compensation model. A continuous correction function for UAV altitude is proposed as a pre-processing step for the model input to compensate for the data, and the complete temperature compensation model is tested and field-validated. The results show that the discrete compensation effect of the GWO-double-layer GRNN network is superior to other models in data testing, with MAE ≤ 0.008 1, RMSE ≤ 0.013 2, and R² ≥ 0.996 9, indicating that the model has good compensation effects. The continuous correction function avoids the influence of UAV altitude on thermal infrared monitoring results, dividing the problem of UAV continuous altitude thermal infrared temperature regression into a stepwise regression calculation, and the final model has a good monitoring accuracy, improving the generalization ability of the temperature compensation model. It provides a supporting calculation method for applying UAV thermal infrared monitoring results to delineate shallow coal fire hazard areas and further promote this method to the corresponding UAV applications and laser monitoring industry.