Abstract: With the large-scale integration of distributed energy resources（DERs） and flexible power loads, the distribution networks are transforming into active distribution networks（ADNs）. This transformation poses significant challenges to the energy management and operation control: 1) The integration of massive DERs requires additional scheduling capacity, which necessitates the active control of DERs to enhance system support capabilities. The variability of these sources also significantly increases the operation risk of ADNs. 2) The complexity and frequent changes in DERs make timely maintenance impractical, and the accuracy of the distribution network model is poor. The engineering application of the precise modeling based operation control and optimal scheduling technology is difficult. To address these challenges, this paper introduces the theory and methods based on machine learning, proposes an energy management and operation control technology system for ADNs that integrates “measurementidentification-control”, and realizes operation control and optimal scheduling with minimal/no model maintenance. Meanwhile, the following key techniques are analyzed: 1) the real-time scheduling technology for distribution networks with weak models or without models, achieving autonomous optimization; 2) the adaptive dynamic control technology for DER clusters, enabling proactive support for the power grid; 3) the probability optimization scheduling method for risk quantification, achieving a balance between risk and economy. Finally, the architecture of the energy management and operation control system suitable for ADNs with a extremely high proportion of DERs is briefly introduced.