Solar PV: Can it Scale Up? A View From the Grid…

I was out at a solar PV site on Friday and talked to some industry folks. One of the big problems with PV is that it doesn’t have any spool down time. This presents a problem because there aren’t any other type of generation sources that can match this quick generation volatility, except for a large scale (expensive) battery.

The problem can be distilled simply: if you have 1MW of PV generation, and the sun goes behind a cloud, how quickly does that 1MW go down, and how far does it go down? My experience with solar PV shows that the time is nearly instantaneous – when the panel is shaded, the output drops immediately. The how far down part depends on the type of cloud. If its low, dark storm clouds, the panels go from nominal output to below the inverter’s threshold of power generation instantly and no power will be generated.

Think of this from the utility’s side – they have 500MW of load to match with 400MW from a natural gas generation facility that can output up to 600MW, and 100MW from solar PV panels that are next to the natural gas plant. If the panels get shaded over the course of a minute, the natural gas plant cant just turn up from 400MW to 500MW immediately. This is due to the nature of the combined cycle generation – modern units have spin-up rates around 7% per minute. So 7% of 600MW is 42MW, so it will take a little over 2 minutes to get back to 500MW of generation. By then the cloud could be gone.

Even if you had software to control the turbines to match the per-second output of the PV array, if the array managed to get shaded faster than the rate the combined cycle system can spin up additional power, then you’re going to have issues matching the load with the power generation.

There is one key to this situation though – how fast can the panels get shaded. If you have a cloud moving overhead at 100 feet per second, and the cloud is 1000 feet long, thats roughly 10 seconds of shade and a 1000′ long shadow on the ground (note I’ve made some assumptions on a few things, like the sun is pretty much stationary over a 10 second period). There are a few things that follow from this…

• Does the turbine get more wear and tear from spinning up at its fastest ramp rate for 10 seconds, the going back down?
• How much PV output gets shaded? Even at 10,000′ long, depending on cloud shape and plant layout, it might only be shading a portion of the total panels, leaving other panels producing at nominal output.
• Does a larger geographic distribution of panels provide for better coverage against the occasional cloud?

There are many questions that need answering when it comes to distributed PV generation. One idea I’ve had is look-ahead demand shaving. In other words, you’d have to build some sort of system to look at the sun and see if there was a cloud coming soon from the perspective of the PV plants. Then, based on the color of the cloud, get an idea how much output will fall, and then send that back to the grid operators. A running net total of PV increases and decreases for the system would be calculated every few seconds so the grid operators would be able to know if the PV-based power output is about to go down, or spike up after the sun comes back out from behind the clouds. They could either use this data to unlock additional power (say, from a flywheel or battery storage system) or send out a message to grid users to try and shave their demand for the next 15 seconds (or whatever time frame is calculated). The backend calculations could get very complex – knowing what data is coming from what geographical position and create a virtual cloud map of the skies above the city, and even doing its own predictions for when dips and spikes will happen.

The problem boils down to if you can geographically distribute the generation, what kind of reliability issues are there on intermittent cloudy days, and how much of that power generation drop can you predict ahead of time to schedule additional generation resources.

I should note wind faces a similar problem. I was looking at a study for how 15GW of wind power in Texas would affect the ERCOT grid. It turns out the maxium 30-minute decrease was around 2.8GW (93MW/minute), which would happen once every 3-4 years, and 2.4GW about once per year. ERCOT, which has a peak demand of about 65-70GW. To accomidate the 2.8GW 30-minute decrease at a 7% ramp rate for combined cycle, you would need to have that much in spinning reserve because even fast starting plants will take about 30 minutes to start. If a plant is at 40% (minimum operational capacity) and had to spin up to 100%, at 7%/min thats less than 10 minutes. So its completely possible to have sufficient reserve within a dispatch pattern, given enough individual turbines (approximately 5GW in the previous example). Also, the report notes that this condition is limited to certain seasonal patterns, so its likely that the dispatch configuration needed would only be required in certain situations.