Fuzzy information granulation-based interval forecasting for fiber quality trends in the refining process

The quality of refining fibers significantly influences the properties of medium-density fiberboard (MDF). Accurate evaluation of this quality is challenging due to the fuzziness, complexity, and nonlinearity of the data, which often cause traditional deterministic prediction methods to fall short. To overcome these limitations, this paper proposes a Fuzzy Information Granulation (FIG)-based interval forecasting model for fiber quality trends. The model employs FIG to convert raw data into granular sequences, mitigating fluctuations and improving feature representation. These sequences are used to train a Support Vector Machine (SVM) model, whose hyperparameters are optimized by an Improved Particle Swarm Optimization (IPSO) algorithm, making it particularly effective for small-sample learning and trend capture. Validation with industrial data confirms that the FIG–IPSO–SVM model excels in short-term forecasting, achieving perfect Prediction Interval Coverage Probability (PICP: 100%) and the narrowest Mean Prediction Interval Width (MPIW: 3.1088), which is 6.38% to 67.20% lower than benchmark models. Despite an expected decline in multi-step forecasting performance, the model offers a practical and effective tool for short-term, real-time quality monitoring and control in the refining process.