Modern power distribution network is becoming increasingly challenging to maintain stable voltage profile, due to increased penetration of renewable energy sources, varying load pattern and limitation of traditional control mechanism. One recent literature presented an intelligent load shedding for voltage control application, it is essentially a data driven based framework and was designed to work in real-time and changeable environment. In contrast to traditional under-frequency or rule-based shedding that is based on static thresholds or offline models, the algorithm presented uses machine learning methodologies to make use of historical and real-time grid state information to predict abnormalities and best load curtailment. Contextual examination will analyse critical and non-critical loads adhering to the least service downtime, with improved reliability. It also incorporates predictive control and renewable generation uncertainty, providing a proactive approach that overcomes one of the main limitations of previous work, that is limited scalability and lacking of flexibility and renewable integration. The experimental results based on a simulated distribution test system illustrate that the proposed model is better not only in terms of keeping voltage stability but also in saving energy and realizing intelligent automation. This paper provides a scalable, intelligent and efficient solution towards the future smart grid modernization.