Trend prediction of gas concentration based on interpolation trapezoidal fuzzy information granulatio

Abstract

For problems that high-level information granules constructed by existing fuzzy information granulation method cannot completely contain information in low-level data, and range of prediction time is limited, an interpolation trapezoidal fuzzy information granulation method was proposed to predict gas concentration trend. Original gas concentration time series is discretized to form sub-series, the maximum and minimum value of each sub-series window are calculated to form trapezoid-top boundary, each sub-series window data is calculated by interpolation to form a new gas concentration time series window, then trapezoid-bottom boundary is calculated by use of data traversing optimization to the new gas concentration time series window, so as to form gas concentration granulation interval sequence. In view of problem that existing evaluation method cannot evaluate effect of information granulation accurately, a granulation evaluation method based on weight value was proposed, which evaluate effect of granulation through the weighted root mean square error. The experimental results shows that granulation effect of the proposed method is better than the existing fuzzy information granulation method, and the granulation effect does not decrease accompany with increase of granulation window, which has high stability and robustness.

Keywords

• trend prediction of gas concentration
• interpolation trapezoidal fuzzy information granulation
• time series
• granulation evaluation
• weighting