With the rapid development of new energy power systems, factors affecting electricity prices have become increasingly complex and diverse, necessitating precise forecasting within intervals of significant price fluctuations. This paper proposes an interval electricity price ultra-short-term forecasting algorithm based on MIC-LSSVM combined with ABKDE. The algorithm first employs MIC to select features with significant mutual information with ultra-short-term electricity prices. Furthermore, it addresses the poor stability of SVM by minimizing the quadratic loss function based on SVM. Simultaneously, ABKDE mitigates the drawback of traditional KDE’s excessive dependence on bandwidth by adaptively adjusting bandwidth according to the local density of data points, thereby providing more refined estimates in areas of drastic density changes. The model’s generalization ability is evaluated through metrics such as PICP and CRPS. Finally, the proposed algorithm is compared with existing ones and validated across different time dimensions, further demonstrating the rationality and superiority of interval prediction.