Establishing a more complete forecasting system of industrial overcapacity risk will help to achieve scientific prevention and precise control of overcapacity, as well as promote high-quality economic growth. Unlike previous literature, we have proposed a new set of forecasting indicator and model systems for coal power overcapacity risk (CPOR) based on the perspective of industrial linkage and the idea of data-driven integrated modeling. First, grounded in industrial linkage theory, we included the upstream, downstream, complementary and alternative industries in a framework of the forecasting indicator system (FIS) for CPOR. Next, we used the filtering and association rule algorithm for dual feature selection of the forecasting variables, and we obtained an FIS of comprehension and emphasis. Second, due to the data’s high dimensionality and sparseness, the cost sensitivity of decision problems, and the machine learning model’s lower interpretability, we built a forecasting model system that covers “model construction → model evaluation → model interpretation”. The empirical results show that our risk forecasting system effectively concerns the accuracy, expected losses, and reliability of forecasting outcomes. Further, we reveal the multi-source inducement of China’s CPOR, identify the key overcapacity risk indicators under different risk levels, and explain the evolutionary law of the risk state. The findings provide comprehensive quantitative analytical tools and a thorough solution for the dynamic monitoring and forecasting of CPOR, as well as a reference and inspiration for other industries.

 

data-driven; industrial linkage; overcapacity; risk forecasting; coal power industry