I propose a new ASME standard, the Scaling Banana™.
They don't float, they're fixed in place at depth. They use the pressure of the surrounding water to spin a turbine as its pumped in and out, the only moving parts are the turbine and its associated components. And seeing as how the water is pumped in and out, most of the silt/detritus pulled in during filling, would be pumped out during draining assuming a siphon tube is used to draw the water from the bottom of the sphere (where all the debris settles) to the pump.
Yes salt water is corrosive, but that problem is already solved, there are currently concrete oil platforms built in the 70s and still in service today. We have formulas for concrete that are proven to be seawater resistant.
Building storage tanks on land wouldn't be as efficient due to the greater pressure differential at 500m underwater vs on land. Dams are one of the most expensive structures to build and are very damaging to the surrounding environment. They also have a much larger problem with silt deposition as there is a constant flow of it, every time it rains there's another surge of silt making its way downstream to be trapped by the dam.
Overall this project would be considerably cheaper, more friendly to the environment, and most likely more efficient than any pumped storage on land. And its not like the sea floor is lacking for real estate, unlike any feasable locations for dams here on land.
You clearly don't understand the basic principles of engineering.