Abstract: In order to achieve accurate prediction of demand for power materials, based on the perspective of misclassification cost, an AdaBoost regression prediction model for agile supply chain demand of power materials is constructed. In the power supply chain, data has characteristics such as strong nonlinearity, high dimensionality, and imbalanced cost of misclassification. Based on this characteristic and considering the cost of misclassification, centralized processing of demand data in the power material supply chain is carried out; design an agile requirement ensemble learning model based on AdaBoost regression, which integrates weights of multiple basic prediction values to improve the predictive ability for high volatility and high cost requirements; by iteratively weighted learning of sample weights, adjusting the weights of samples, and integrating a combination of weak learners, the optimization of the prediction model is achieved. The comparative experimental results show that the designed method can not only achieve the prediction of agile supply chain demand for power materials, but also control the training loss of the prediction model while reducing the deviation of the prediction results, ensuring the high reliability of the prediction results.