Metro Turnout Short-term Action Overload Early-warning Method Based on Multi-description Complementary Prediction

Abstract

[Objective] Metro turnback turnouts undergo frequent movements and complex working conditions, making them prone to overload actions that lead to mechanical failures such as jamming and blockage. A method for predicting short-term action loads and assessing overload early-warnings is proposed. [Method] Focusing on metro turnout short-term action overload early-warning, research is carried out expounding the assessment method for turnout short-term state early-warnings. Utilizing the proposed multi-description weighted prediction mechanism and adaptive kernel density estimation, the approach to achieve interval prediction of action loads and the calculation of early-warning assessment indicators are described, and the early-warning assessment process is outlined. Based on this, an application plan for action overload early-warning is established and empirically tested using turnout action data from Wuhan Metro. [Result & Conclusion] First, real-time continuous collection of turnout power curve characteristic points is conducted, and the root mean square value is calculated to obtain the characteristic sequence of turnout action loads as the input for prediction. Then, a multi-description complementary prediction mechanism is designed, using ELM (extreme learning machine) as the base predictor to establish a multi-description complementary prediction machine to carry out multi-step range prediction of short-term action load characteristics, forming a prediction feature set. Lastly, taking the elements of the prediction feature set as samples, the probability density function and the overall value confidence interval of the turnout action load characteristics during the prediction period is calculated using an adaptive kernel density estimation method. Combining this with the confidence interval of historical load characteristics, the action overload degree during the prediction period is defined as the early-warning assessment indicator. Empirical test results show that for pre-fault load characteristic data, the coverage rate of the estimated confidence interval by prediction reaches 94.2%, and the confidence interval width generally matches the actual value interval width changes. When the early-warning assessment threshold is set at 0.63, all test cases can successfully issue early-warnings during the 4th to 9th actions before turnout faults occur. The test results validate the effectiveness of the overload early-warning method.

Keywords

• metro
• turnout
• overload early-warning
• multi-description complementary prediction
• adaptive kernel density estimation