This paper proposes a reconstructive machine learning approach for detecting distributed denial of service (DDoS) attacks in smart grid networks. The method uses deep and shallow reconstructive models to learn representations for each attack type and performs attack detection through class-specific reconstruction error-based classification. The key advantage of this approach is the ability to add new attack classes without retraining the entire model, minimizing disruption to smart grid operations. The technique was evaluated using two standard DDoS attack datasets and compared against six existing methods. Results showed that the proposed method achieved higher accuracy and did not require full model retraining when new attack classes were introduced. The method enhances the security and stability of smart grid networks by effectively detecting DDoS attacks, protecting critical infrastructure from evolving cyber threats. The approach leverages various reconstructive models, including deep autoencoders, marginalized stacked denoising autoencoders, and extreme learning machine autoencoders, to extract meaningful features from network data. Experimental results demonstrated the effectiveness of the proposed method in accurately classifying DDoS attacks, with high precision, recall, and F1 scores across different datasets. The method's ability to adapt to new attack types without retraining makes it a robust and efficient solution for DDoS detection in smart grid networks.This paper proposes a reconstructive machine learning approach for detecting distributed denial of service (DDoS) attacks in smart grid networks. The method uses deep and shallow reconstructive models to learn representations for each attack type and performs attack detection through class-specific reconstruction error-based classification. The key advantage of this approach is the ability to add new attack classes without retraining the entire model, minimizing disruption to smart grid operations. The technique was evaluated using two standard DDoS attack datasets and compared against six existing methods. Results showed that the proposed method achieved higher accuracy and did not require full model retraining when new attack classes were introduced. The method enhances the security and stability of smart grid networks by effectively detecting DDoS attacks, protecting critical infrastructure from evolving cyber threats. The approach leverages various reconstructive models, including deep autoencoders, marginalized stacked denoising autoencoders, and extreme learning machine autoencoders, to extract meaningful features from network data. Experimental results demonstrated the effectiveness of the proposed method in accurately classifying DDoS attacks, with high precision, recall, and F1 scores across different datasets. The method's ability to adapt to new attack types without retraining makes it a robust and efficient solution for DDoS detection in smart grid networks.