Prediction model of coal spontaneous combustion oxidation state based on MICPO-LSSVM
•A novel model combining game theory and grey target decision determines the oxidation state of coal spontaneous combustion.•A multi-strategy improved CPO algorithm is proposed, significantly enhancing its performance.•A prediction model for coal spontaneous combustion is established and validated e...
Uloženo v:
| Vydáno v: | Fuel (Guildford) Ročník 407; s. 137366 |
|---|---|
| Hlavní autoři: | , , , , , |
| Médium: | Journal Article |
| Jazyk: | angličtina |
| Vydáno: |
Elsevier Ltd
01.03.2026
|
| Témata: | |
| ISSN: | 0016-2361 |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| Shrnutí: | •A novel model combining game theory and grey target decision determines the oxidation state of coal spontaneous combustion.•A multi-strategy improved CPO algorithm is proposed, significantly enhancing its performance.•A prediction model for coal spontaneous combustion is established and validated experimentally.
To achieve more accurate prediction of coal spontaneous combustion (CSC) temperature and oxidation state, a Least Squares Support Vector Machine (LSSVM) optimized by the Multi-strategy Improved Crested Porcupine Optimizer (MICPO) was developed. First, programmed temperature-raising experiments and thermogravimetric analysis of CSC were conducted. Based on the experimental results, predictive indicators for the oxidation state were selected, and the oxidation stages of CSC were defined. Subsequently, the Improved CRITIC (ICRITIC) method and Improved Analytic Hierarchy Process (IAHP) method were employed to calculate the weights of the indicators, and a game theory combination weighting method was applied to optimize these weights. Then, based on the weighted grey target decision principle, a comprehensive target distance distribution set corresponding to different CSC temperatures was established and used as the input features for the model. MICPO was used to optimize the LSSVM, thereby improving the prediction accuracy of the model. A comparative analysis of the prediction results from the MICPO-LSSVM, Sparrow Search Algorithm-Back Propagation Neural Network (SSA-BPNN), Whale Optimization Algorithm-Bidirectional Long Short-Term Memory (WOA-BiLSTM), and Particle Swarm Optimization-Support Vector Machine (PSO-SVM) models showed that: in terms of regression performance, the MICPO-LSSVM model reduced the Mean Absolute Error (MAE) by 12.22, 11.79, and 5.63; the Mean Absolute Percentage Error (MAPE) by 4.82%, 4.58%, and 1.48%; the Root Mean Square Error (RMSE) by 16.45, 16.52, and 24.32; and increased the Coefficient of Determination (R2) by 0.06, 0.06, and 0.10, respectively. In terms of classification performance, the model improved Accuracy by 0.15, 0.21, and 0.17; Macro-Precision by 0.34, 0.30, and 0.31; Macro-Recall by 0.23, 0.30, and 0.27; and Macro-F1 by 0.29, 0.30, and 0.31, respectively. The MICPO-LSSVM model was applied to predict the oxidation state of CSC in the goaf of the 1303 fully mechanized top-coal caving face at Jinniu Coal Mine. The prediction results were consistent with practical engineering conditions, demonstrating that the MICPO-LSSVM model is suitable for predicting the temperature and oxidation state of CSC. |
|---|---|
| ISSN: | 0016-2361 |
| DOI: | 10.1016/j.fuel.2025.137366 |