本文采用特征选择的新技术，以“理论研究—实验数据应用—实际数据应用—结论及分析”为设计路线，对矿山监测监控系统中测量的数据进行了处理，为煤矿预测预报系统提供可靠的数据基础，从而达到提高我国煤矿安全生产能力的目的。 论文选用遗传算法来实现特征数据的选取，并应用MATLAB遗传算法工具箱中的工具函数编程实现。主要完成的工作和结论如下： 首先，将数据根据选取的特征数据量的大小分段，对遗传算法的编码方法、选择、交叉和变异操作做了比较选择，在段内单独运行遗传算法，选取其结果作为该段的特征数据，循环执行完所有数据段后，即得所需的特征数据。充分利用了遗传算法在全局搜索和优化方面的性能。 其次，以实验数据和实测数据来分别验证本文算法。从正弦函数叠加产生的白噪声波形中获得实验数据，分别以二进制编码和浮点数编码验证了算法的可行性；针对实测数据具有数据量大、取值范围广、精度要求高的特点，为提高遗传算法的性能，首先对原始数据进行预处理，然后采用浮点数编码，通过对分段内的原始数据增减一个较小数据值来构成初始种群的取值范围。 最后，无论是实验数据还是实际数据的运行结果，都表明本文的特征选择算法选取出来的特征数据，能够充分代表原始数据，达到本文的要求；同时，还得出了关于改变终止代数可以随之改变提取率的结论。另外，本文的算法还具有一般性，可用于其他监测监控系统的特征数据选择中。

According to the thought of “theory research, experiment data validate, actual data application, conclusion and analysis”, and new technology of feature selection, the acquired data which reflects the safety state in coal mine is processed, and that improve the safety state of coal mine in China because it provides credible data for safety prediction. Genetic Algorithm (GA) is applied to select the characteristic data, and the realization of GA is based on the functions provided by MATLAB GA tools. The following show the primary methods and conclusions in this paper. Firstly, the data is separated to subsections according to the number of the characteristic data needed. Using GA in subsections and then we can adopt the results as the characteristic data of subsections. This has taken full advantage of global searching and optimization performances of GA. Secondly, both the experiment data and actual data are used to validate the algorithm. According the experiment data acquired from sine wave with white noise, the binary code method and floating-point code method are used to validate the feasibility of algorithm. Considering the characteristics of actual data such as great capacity, wide value scope and high precious, the follow method is applied to improve the performance of GA. After pre-processing of original data and floating-point coding, the value scope of initial population can be obtained by adding or reducing the original data in subsections. At last, both the validating results of experiment data and actual data indicate that: firstly, the characteristic data obtained by the feature selection algorithm represent the original data sufficiently, and that satisfy the desire of paper; secondly, the extraction rate changes following the termination generation. Furthermore, the algorithm can be used in other monitoring systems to acquire data feature.