Introduction  Under the constraint of lean investment in power construction, precise cost control of substation engineering has become an increasingly concerned issue for power suppliers. To address current difficulties in cost prediction due to the large number and great complexity of influencing factors, this paper proposes a substation engineering cost prediction model based on machine learning algorithms.

  Method  Firstly, important influencing factors were selected from historical substation construction cost data using methods such as the analytic hierarchy process, analysis of typical projects, the Delphi method, and the correlation coefficient. Relevant data on substation engineering costs were collected through extensive investigation to form a substantial training dataset for model validation and testing. Then, key parameters in the Support Vector Regression (SVR) model were optimized using cross-validation and the Bayesian optimization algorithm to minimize prediction errors. Finally, the optimized SVR model was used for cost prediction, and an empirical validation was conducted.

  Result  The results show that the SVR model not only demonstrates a robust fit to the training data but also excels in generalizability. It achieves good accuracy in predicting the total settlement prices of substation engineering costs as well as the costs of various sub-projects.

  Conclusion  This approach enables scientific forecasting and effective management of construction costs during the substation design phase. It can offer a methodological reference for precise cost predictions in substation engineering projects.