Considering the electricity market, data analytics paves the way for completely new strategies regarding demand and supply-side policies. In this manner, predictive analysis of the demanded power accuracy is carried out to boost profits and increase the penetration of similar demand response (DR) programs across all levels of end-user categories. Residential loads experience stiff spikes and unpredictable variations due to occupancy activities and environmental factors. To address this, we first propose a robust short-term multivariate-multistep forecasting framework that is resilient to missing or erroneous data, employing temporal convolution networks (TCNs). We then incorporate two distinct valley-filling indices to optimize the charging of electric vehicle loads according to DR requirements, showcasing the efficacy of leveraging artificial intelligence to enhance the utilization of clean energy resources. Simulation studies are conducted using real-world nodal residential loads with hourly granularity. The results demonstrate that the forecasting method is reliable for residential locations, even when dealing with highly damaged data. The case studies effectively fill the load into the valleys and minimize fluctuations in residential locations. Through the integration of emission-aware forecasting and optimization strategies, our study lays the groundwork for a comprehensive approach that not only improves economic outcomes and grid stability but also advances the imperative of reducing carbon emissions.