Research on IPSO-SVM silicon material supply and demand early warning method based on supply chain perspective

With the rapid growth of the photovoltaic industry, the instability of the silicon market has led to an imbalance in silicon supply and demand. To address this problem, this paper proposes an early-warning method for silicon supply/demand based on the improved particle swarm algorithm optimized support vector machine (IPSO-SVM). First, based on a supply chain perspective from the supply and demand sides using time-difference correlation analysis, the impact of silicon supply/demand balance on five warning indicators was determined, as well as a benchmark indicator of seven warning levels and warning thresholds. The silicon supply/demand warning model, based on the IPSO-SVM, suffers from the drawbacks of the PSO algorithm, including a tendency to converge to local optima and slow convergence. To address these limitations, a combined optimization strategy was applied to the PSO and integrated within the IPSO-SVM model to achieve self-adaptive parameter tuning. Warning indicators were used as model inputs, and the warning level of the benchmark indicators served as the model output, enabling the model to predict the category of the warning level. Finally, this study introduces an enterprise X dataset for validation. The results show that, compared with a PSO-SVM model with single-strategy optimization, IPSO-SVM achieves an average accuracy improvement of 3.51%. Compared with the standard PSO-SVM model, IPSO-SVM achieves an average accuracy improvement of 4.35%. Moreover, relative to SVM classification models optimized by various other algorithms, IPSO-SVM attains an average accuracy gain of 12.28%, with a notable improvement of 7.45% for the minority class, thereby significantly enhancing the overall performance of the classification model.