A neural tuner is used to increase energy efficiency of unsymmetrical plants described by first or second order aperiodic links with time delay. It allows tuning Kp and Ki parameters of a Pi-controller online without knowledge of a plant model. A main part of the tuner is a neural network, which input vector includes plant output value signals delayed on equal time gaps from each other. The main aim of the research is to develop a method to calculate an optimal value of such time gap, i.e. to find a dependence between time gap value and plant parameters values. More than 15000 experiments are conducted with plant models using different values of time constant, plant gain and delay time. Time gap value is changed from 1 second to 40 seconds for each certain model. The best experiment is chosen from the each set of 40 on the basis of proposed criteria. Such an experiment shows the best value of time gap for the plant model in question. Having conducted experiments, sought analytical dependence is found. It is also shown that the number of the neural tuner calls N during each transient of each experiment with the best value of time gap is a constant. N value probability curve has a Gaussian distribution. On that basis, a method to calculate time gap value without knowledge of plant parameters values is proposed. Further research needs to be done to include time gap parameter into neural network on-line training process to be able to refine it during control system functioning. In that case, obtained analytical dependence will be used to initialize time gap parameter.