<p>Most existing constrained multi-objective evolutionary algorithms are not so efficient when handling constrained large-scale multi-objective problems (CLSMOPs). To overcome white-box CLSMOPs with definitive objective functions, a spatial–temporal lever evolutionary algorithm (STLEA), consisting of the lever evolutionary algorithm (LEA) and spatial–temporal preference strategy (STPS), is proposed. LEA ditches the thought of the mainstream algorithms for the similar problems, which changes the structure of the large-scale decision space, but handles the large-scale decision space by a certain small-scale decision space. Specifically, inspired by the lever principle, LEA explores the way to pry up the large-scale decision space, as the “load”, by the small-scale decision space, as the “force”. Meanwhile, LEA rotates the optimizations in between “load” and “force” for dual-balance: balance between objectives and constraints, and balance between convergence and diversity of solutions. STPS dynamically adjusts the proportion of optimizations in “load” and “force”. Different from existing preference strategies, which only consider the stage of the evolutionary procedure, STPS considers both stage, related to time, and varying scale of the decision space, related to space, for the comprehensive balance of feasibility, convergence, and diversity of solutions. Eleven representative and state-of-the-art constrained multi-objective evolutionary algorithms have been compared to the proposed STLEA to demonstrate its effectiveness through comparative experiments on through comparative experiments on CLSMOPs with equality and inequality constraints and 1000 decision variables and three typical MaMIMO-LU models with 1024 antennas and 128, 256, and 512 users. Experimental results show that STLEA achieves the best SE-EE tradeoff.</p>