关键词: 瓦斯浓度, 相空间, 时间延迟, 嵌入维, 加权一阶局域法

Abstract: A prediction model of non-linear time series based on K-Means clustering was proposed. Using the ability of short-term predicting for chaotic time series, the paper constructed a gas concentration prediction model for certain coal mines. Correlation integral method was used to determine the time delay τ and dimension m. After the phase space was reconstructed, the weighted one-rank local-region method based on K-Means clustering was used to construct prediction model. The experimental results show that, if next one minute data will be forecasted, it is more appropriate to use a continuous 200 training data to determine parameters τ and m for predicting better results, the error reaches 0.0341; if next few minutes data will be forecasted, it is more appropriate to use a 200 training data with 15 minutes intervals for predicting better results, the error is 0.0437. It shows that the timing of the gas concentration restores the initial chaos after 15 minutes.

Key words: gas concentration, phase space, time delay, embedded dimension, weighted one-rank local-region method