Abstract: The performance degradation of airborne fuel pump is nonlinear and multi-stage degradation pattern. In order to improve the prediction accuracy of the fuel pump performance degradation and get an accuracy prediction range, a novel prediction method based on Singular Value Decomposition-fuzzy information granulation and optimized Extreme Learning Machine (ELM) is proposed. Aiming at the defect of the traditional granulation prediction method which directly use the granulation analysis to the original series, Singular Value Decomposition is executed for those degradation index sequences and the trend term and detrend term are obtained. Fuzzy information granulation is executed for the de-trend terms. Then, the trend term and the granulated detrend term are input into the ELM to perform regression prediction. In this process, particle swarm optimization (PSO) is used to optimize ELM parameters. Finally, the prediction model is evaluated according to the comparison between the measured values and the predicted values. Experimental results show that the proposed method can effectively rack the trend and realize the change trend and spatial prediction for the fuel pump performance degradation index.

Key words: Fuel pump;Parameter prediction;Fuzzy information granulation;Singular value decomposition;Extreme learning machine

摘要： 机载燃油泵的性能退化呈现非线性多阶段模式，为了提高机载燃油泵性能退化指标的预测精度，得到性能退化指标准确的预测范围，提出了基于奇异值分解-模糊信息粒化与优化极限学习机的模糊粒化预测方法。针对传统的粒化预测方法直接对原始序列进行粒化分析的不足，首先利用奇异值趋势分解方法提取燃油泵性能退化指标序列的趋势项及去趋势项，再利用信息粒化方法对去趋势项进行模糊粒化；然后将趋势项及粒化后的去趋势项数据输入至极限学习机进行回归预测，并采用粒子群算法优化极限学习机参数；最后根据实测值和预测值的对比分析评估预测模型的优良性。实验结果表明，该方法可以有效跟踪燃油泵性能退化指标的变化趋势，并对其指标的波动范围进行有效预测。

关键词: 燃油泵;参数预测;模糊信息粒化;奇异值分解;极限学习机