A New Cost Estimation Model for Small-scale Grid Infrastructure Projects based on CNN and SVM

Over the past few years, there has been a progressive upsurge in the quantity of infrastructure with small scale initiatives within the electrical grid. Predicting the cost of such projects can effectively control construction fees and reduce investment risks. This paper integrates convolutional neural networks and support vector machine to construct a CNN-SVMbased prediction model, which not only leverages the strengths of both CNNs and SVMs but also overcomes their respective limitations, thereby achieving superior predictive performance. In order to enhance the model’s ability to optimize parameters and to prevent it from falling into local minimums, northern goshawk optimization algorithm is introduced to train the penalty parameter and kernel function parameter within the model. Through a numerical experiment, the proposed model is demonstrated to perform better than the single model.