Discriminating water inrush sources efficiently and accurately is necessary to control water in coal mines. We combined the improved genetic algorithm (IGA) and extreme learning machine (ELM) methods and applied this new method to the Zhaogezhuang mining area. The IGA-ELM method effectively solved the complex non-linear problems encountered in identifying water sources and proved to have several advantages over conventional methodology. The IGA for the hill-climbing method was adopted to use the weights and thresholds of the ELM, which overcame the prematurity of the traditional genetic algorithm and the instability of the ELM model. Three types of water were identified in different aquifers of the Zhaogezhuang mining area: SO₄-Ca in the Laotang water, SO₄·HCO₃-Ca in the Ordovician limestone water, and HCO₃-Ca in the fractured sandstone roof of the no. 12 and 13 coal seams. The water sample recognition was 95% accurate, which proved that the water inrush source in the Zhaogezhuang mining area was accurately identified by the IGA-ELM model.

有效、准确地判别突水源是控制煤矿用水的必要条件。我们将改进遗传算法（IGA）和极限学习机（ELM）方法相结合，并将这种新方法应用于赵各庄矿区。IGA-ELM 方法有效地解决了识别水源时遇到的复杂非线性问题，并证明与传统方法相比具有多个优势。爬山法的IGA采用了ELM的权值和阈值，克服了传统遗传算法的早熟和ELM模型的不稳定性。赵各庄矿区不同含水层中鉴定出三种水：老塘水中的SO ₄ -Ca、SO ₄ ·HCO ₃-Ca 为奥陶系石灰岩水体，HCO ₃ -Ca 为第 1 号裂隙砂岩顶板。12 和 13 煤层。水样识别准确率为95%，证明IGA-ELM模型准确识别了赵各庄矿区突水源。