According to this website, single Bitcoin transaction footprint is:
Their overall power calculations are indeed pretty accurate.
In fact, anyone could approximate the electricity consumption simply by taking the hashrate, dividing by the most efficient gear that exists, and multiplying the number of gears required by their consumption, then you have the minimum, there is no way to go below those unless we go into conspiracy theories about entities using far more efficient gear or manipulating the algorithm.
But the consumption per transaction is wrong because there is no real accurate model to calculate this.
The simple way is to grab the power consumption and split it into the number of transactions but what happens when we have variables like the last week as onn the 3rd we had only 220k, on the 6th 330k, but the power consumption didn't go 50% up.
The second problem is the transaction itself, you can have one transaction with 60 outputs, which means 60 people received their payments, you can have one transaction in which you only pay for yourself, and you can have a transaction with 100 inputs, larger than both in which you just consolidate your dust.
And it doesn't stop here, there is the empty block problem and there is of course the LN network which is never taken into consideration.