In order to reduce the network loss of an active distribution network and improve the reliability of system operation, a mathematical model with the goal of network loss, switching operations, load and voltage balance is established. Given the problems of poor global data dependence of traditional harmony search algorithms, slower speed in the second half of optimization, and easy falling into a premature dilemma, the adaptive parameters, value mechanism and out-of-bounds escape are improved. Random optimization compensation for reactive power, using scenario analysis to determine the output value of the wind turbine. It also uses the electric vehicle model in timing mode to cut peaks and fill valleys for daily load, and the K-Means method based on Euclidean distance segmenting the distribution network containing distributed power sources within a day. It carries out a dynamic and static double test in the IEEE33 power structure. Compared with traditional harmony and adaptive harmony search algorithms in terms of algorithm convergence, it reduces the number of invalid runs of the algorithm and improves the running speed of the algorithm and the global optimization capability. Compared with the improved ant colony algorithm in the reconstruction optimization results, the network loss and the number of switching actions are reduced, and the economy and reliability of the system operation are improved. This work is supported by the National Natural Science Foundation of China (No. 61672337).