The construction of a lean operation and inspection integrated management system for substations is an important part of the development and maintenance of the power system. Forecasting the investment benefits of substation project development is an important issue in feasibility analysis. Therefore, we need to use a highly accurate method to make a prediction of the investment benefit of this project. Granger causation is a causal relationship based on “prediction”, and inferring about its causality is a key task in time series analysis. In this paper, we propose a new estimation method, Granger causality estimation based on supervised learning. This method uses an eigenvalue representation of the distance between conditional distributions conditioned on past values. And for different time series, the method can give different feature vectors. Applying it to the prediction of the investment efficiency of the substation can achieve a good prediction effect. Therefore, we used granger causality to build a predictive model of the return on investment in substations.