Robust Interval Prediction of Intermittent Demand for Spare Parts Based on Tensor Optimization

Demand for spare parts, which is triggered by element failure, project schedule and reliability demand, etc., a kind of sensing data to the aftermarket service large manufacturing enterprises. Prediction demand parts plays crucial role in inventory management lifecycle quality large-scale In real-life applications, however, occurs randomly fluctuates greatly, sequence shows obvious intermittent distribution characteristics. Additionally, due factors such as reporting mistakes made personnel or environmental changes, actual are prone abnormal variations. It thus hard capture evolutional pattern traditional time series forecasting methods. The prediction results also reduced. To address these concerns, this paper proposes tensor optimization-based robust interval method aftersales parts. First, using advantages decomposition effectively mine intrinsic information from raw data, sequence-smoothing network based on stacked autoencoder proposed. Tucker applied hidden features encoder, obtained core reconstructed through decoder, allowing us smooth outliers original sequence. An alternating optimization algorithm further designed find optimal feature representation extraction evolutionary trend series. Second, an adaptive with dynamic update mechanism obtain point values intervals sequence, thereby improving forecast. proposed validated engineering enterprise China. experimental demonstrate that, compared typical methods, can grab various improve accuracy predictions small-sample Moreover, provides reliable elastic when distortion results, offering new solution intelligent planning decisions related practical maintenance.