參數數值無法確定是影響系統性能及可靠度的主要原因之一,本研究建立辨識校準參數的流程,以確認運行系統之各參數數值。然而校準參數在複雜系統應用上可能會遇到問題有(1) 參數過多造成校準困難,(2) 參數校準準確率不足,以及(3) 參數校準結果信心水準不足。本研究藉由主因素分析找出系統的重要參數,降低複雜系統的分析難度,根據系統性能偏移,以類神經網路校準參數數值,再利用多個根據不同性能偏移以類神經網路校準參數的結果,以決策樹提升校準準確率,並以信賴區間評估參數的校準結果。研究以一車輛動態測試的工程案例作為演示,車輛參數校準方均根誤差最小可達0.136%,本研究所提出之方法可有效校準偏移之參數,並提供校準複雜系統參數的完整分析流程。
Parameter uncertainty plays an important role in system performance and robustness. This research builds up a procedure for calibrating deviated parameters. However, there may be difficulties applying parameter calibration in complex system, namely (1) computation inefficiency due to a large number of parameters, (2) inaccuracy in parameter calibration, and (3) low confidence in calibration result. This research selects important parameters by main effect analysis and uses the neural network to calibrate parameters via performance deviation. After getting calibration results via different performance deviation, we use the decision tree to increase the accuracy of calibration and evaluate the result by applying confidence interval. The method is demonstrated in an engineering case: vehicle dynamic test, the minimum mean square error of calibration is 0.136%.