Electronic – SMPS : Average Current Control Vs Peak Current Control

In short you need to choose the part to survive the peak current.

Longer Answer:

I'd work your way through the SEPIC topology's design calculations again, your current ripple is really high, the difference between average and peak current should be much smaller.

The datasheet says that ΔIsw should be 20-40% of Isw(average max). Or to use the datasheet's terminology χ should be between 0.2 and 0.4.

In your case (estimating by looking at the simulation output graph):

Isw(average max) = ~2.75A

ΔIsw = ~5.5A

χ = ~2.0

Try rerunning your simulation with larger L1/L2 inductors, this should reduce the ripple currents.

EDIT: try a larger cap for C3 also (try 10uf or a second 4.7uf in parrellel), the ESR / ESL of this part is also critical, check what the values of these are for your simulation model.

I am fortunate enough to have a Bode 100 Vector Network Analizer at work (I work at Omicron Electronics), so I am used to measuring frequency responses of devices and impedance across frequency for many circuits.

Of these, I find resonant circuits the most interesting, but what you are seeing is not totally strange for me. It is very normal to see some of these "non-ideal" effects when you are doing some measurement, and it is normally due to parasitics.

Winding capacitances, parasitic capacitances and inductances in the power stage, and leakage capacitances in general, as well as parasitic inductive couplings between components, and also from the circuit to the chasis or earth, do not normally have an appreciable effect, but when you do a frequency sweep, they sometimes resonate with the components in the circuit and you can have some local smaller resonances which alter the expected "ideal" ac impedance of your circuit.

I'm not really sure that you will have a control loop problem, you could run into local frequency oscillations, but they are local oscillations anyway. But if it really proves to be a problem, what I would do is to quantize the set of frequencies you can choose so that they are enough far apart to ensure monotonicity.

To avoid the output "roughness" that such a quantization would produce, you can recover part of the actuator fine-grained operation by alternating between different frequencies in a pwm-like fashion. I.e: to get the power that would correspond to the midpoint between two frequencies, you could operate half of the time on each frequency.