The only thing is you can know is that which algorithm can avoid local minima. Some of the algorithms can avoid that such as Simulated Annealing. Later you test the algorithms which can avoid the local minima for your application. The best results you get can be assumed the best method. But off course you will need some functions to measure the accuracy of your results.
It is difficult to generalize one optimization technique for all engineering applications. It depends on the problem and the degree of complexity. I have used GA method for a renewable power generation system (specifically solar) to maximize the power and its efficiency which did the job. It has been widely used and recommended by many researchers as a powerful tool for optimizing the engineering problems. However, I advise you to go through the literature to find out the most recommended algorithm suitable for your subject.
I would like to recommend you to apply differential evolution (DE) for solving this problem. DE is fast and give better results compared to may other algorithms mentioned by you. You may use the following article as reference:
Every optimization techniques has its own advantages and collateral disadvantages.
As far as renewable energy problems are concerned, if you are sure about the accuracy of your input data and doesn't want to incorporate any randomness factor (if any) in calculation you can simply go for classical optimization techniques as long as you have enough variables (upto 15, though classical techniques allow to handle upto 100 variables as some say but it will increse your computation load). So at higher no of variables go with meta heuristic approach.
GA and neural networks are good with randomness but we dont generally see it in renewable energy system problems. So for that SA, PSO, Ant Colony will give faster results. Choice is obviously yours, how you define your problem mathematically.
it is highly imposible that any algorithm fit to all problem. Use of optimization method is highly depend on characteristic of problem. If problem is small scale, sqp metgod peform best as my experience. With increasing complexity and nonlinearity, AI method will be more efficient. Further, in AI method, second step of improvisation more effect on the quality of solution. In almost modified algorithm, the second step is modified therefore.