Abstract: Accurate forecasting of rush-repair spare parts demand is essential for enhancing the emergency response capability of power grids and minimizing unnecessary inventory holding costs. The challenge arises from the inherent characteristics of these spare parts, whose application scenarios are highly uncertain, complex, intermittent, and randomly distributed. To address this problem, this study aims to develop an efficient and timely prediction approach capable of capturing both the stochastic and scenario-dependent nature of spare parts demand. This paper proposes a hybrid methodology that integrates scenario evolution analysis with fuzzy inference. The framework consists of two core components: a Bayesian Network (BN) model for scenario prediction and an Adaptive Neuro-Fuzzy Inference System (ANFIS) for quantitative demand forecasting. The BN structure is learned using the Hill-Climbing algorithm combined with the Bayesian Information Criterion (BIC) scoring function, enabling accurate identification of key scenario evolution paths. The methodology proceeds in two stages. (1) Scenario Identification: Critical influencing factors are examined to define decision scenarios, and the BN model is used to predict and recognize typical operational scenarios that trigger spare parts demand. (2) Quantitative Forecasting: ANFIS is employed to extract fuzzy rules from both numerical data and expert knowledge, mapping the predicted scenarios to the corresponding quantitative demand outcomes. The results demonstrate that the proposed method achieves superior forecasting accuracy and robustness compared with benchmark time-series models such as SVR, LSTM, and RF, particularly when dealing with intermittent, sparse, and zero-inflated demand patterns. The effectiveness of the approach is validated through a real-world case study involving a power grid company's rush repair spare parts demand over a three-year period (2019–2021), covering both stable and highly volatile operational conditions. This research provides a novel perspective and a hybrid predictive framework for addressing the complex forecasting challenges associated with highly random and intermittent spare parts demand. It supports a strategic shift in spare parts reserve management from a reactive, “emergency-driven” mode toward a proactive, “prevention-driven” strategy, offering valuable insights for inventory and logistics management in similar high-uncertainty contexts.

Key words: rush repair spare parts, demand forecasting, scenario analysis, fuzzy rules