How could grid-tied Photovoltaic Plants become more competitive when it comes to National Electricity Markets participation? (Part III)
After several weeks, I am back to share with you the third and last part of my debut article (post)! I am sorry for the delay but, meanwhile, we have had the opportunity to enjoy reading some other really interesting posts in the Blog!
So far, we have done a quick review on the concepts that we are dealing with when we talk about Grid-tied Energy Systems + Storage, and we have also described and showed some details about the strategy that is proposed by the authors. Now, it is time to see how our plan works! So, in this post some simulations are analysed in order to present the results of the algorithm when it deals with different meteorological conditions.
One decision to be taken is the way we want to deliver the committed energy to the Electricity Grid during the operation day. In our case, we propose a very simple delivery which will consist of constant values of power, so that at the end of each hour slot the accumulative value of that power curve (integral) will be equal to the energy committed with the Market. In short, applying a constant power of 10 MW during one hour would result on an Energy Supply of 10 MWh, as simple as that. This way of meeting our energy commitments is only possible thanks to the Storage, as it was explained in our first post. Otherwise, our power curve during one day would have the typical shape of an inverted parabola, since we would need to deliver the exact amount of power that our solar panels are producing at any time.
Below, a number of cases are shown, starting from an ideal (and unlikely) situation and ending up with a “bad day” for our PV Plant and our management strategy. The first plots will be explained with more detail, whereas the others will be briefly commented.
Case 1: Perfect Hourly Prediction in a Sunny Day
In this first hypothetical situation, we are able to perfectly predict the weather conditions during the operation day and, therefore, we are supposed to meet the committed energy with the Market. No problem at all.
The graphics generated by the Phase 1 (Scheduler) for this case are shown below. It must be reminded that the first curve (Energy to be provided during the operation day) is made up of the optimal energy levels (from an economic point of view), which have been established taking into account, among others aspects, the available weather prediction.
Not wanting to make the post so tedious, we will just focus on our two outputs (two graphs in the upper half), which are the Power Supply and the Energy Stored in our Battery Farm. It can be seen how the black and the blue curves are overlapped in the first graphic and how the blue and the magenta curves are perfectly matching at the beginning and the end of each hour. That means that the predicted energy is stored at the end of each time slot.
After running the simulation for this first ideal case, the correlation of the results were found to be perfect.Zoom to Graphic 1:
Zoom to Graphic 2:
The errors along the operation day can be automatically calculated and plotted, as it is illustrated below.
In that curve, a positive value means that we have not reached the committed energy level. A negative value, on the other hand, means that an excess of energy has been supplied at a given hour. In this particular case, the maximum errors are:
Defect: 100 Wh
Excess: 228.8 Wh
Which are both null errors when it comes to huge grid-tied Power Plants. We will go faster through the upcoming cases, since we already know the meaning of each plot.
Case 2: Unexpected Cloudy period of time
A really intuitive explanation of the situation is given below.
In terms of Power generation, the main consequence is that lower power production will be available from 1.00 pm to 3.20pm, as it is represented by the red curve in the graph below. This situation is pretty ideal as well, since during almost the whole day real production and prediction have perfectly matched.
In this particular case, due to those clouds we will be generating 10.35 MWh less than expected.
Given this situation, will our real-time predictive controller be able to properly operate the Storage to compensate the lack of solar energy? Let’s find out!
Below, the results of our operation day simulation are shown.
It is observed in the first plot that reference and power delivery are still overlapping. It means that, despite the fact that we have less production, our controller is able to discharge the exact amount of energy during the adverse period of time and avoid, that way, financial penalties.
As a result, the second plot shows how the reference for our Storage is not followed, since we have suffered a non-predicted cloudy period of time.
The errors for each hour slot are illustrated below.
Maximum error of 10e-4 MWh, which is again considered null in bulk energy applications.
Case 3: Unexpected Overproduction
This situation can also be given: our prediction is less optimistic in terms of energy production than the reality. Therefore, we have made our commitments much lower than they could be. So, what are we supposed to do with that extra power generation? Would our strategy be able to handle this case?
Well, thanks to the Energy Storage Systems we can store the overproduction instead of wasting it. It must be reminded that we will have to face financial penalties if we supply more energy than the committed level. The only problem is that we would come to a point when it is physically impossible to keep loading our batteries; situation that will be also analysed in upcoming lines.
The curves that represent this fourth situation are shown below.
Cumulative Overproduction at the end of the Operation Day: 173.36 MWh.
Below, the simulated operation day is presented. Although the overproduction is huge, our controller is capable of identifying that there is enough storage available to keep loading energy. Notice how in the first plot the blue (Power Generation) and the black (Reference to follow) lines overlap. However, it can be seen how the Stored Energy is gradually rising along the day and, as a consequence, the planned reference for that output is not followed at all. This behaviour has a lot to do with the weights inside our predictive algorithm that were commented in Part II.
The maximum error during the operation day is 1.8736e-05 MWh, which is the smallest one so far.
Our fourth and fifth cases represent situations in which there is no physical possibility to avoid the financial penalties mentioned above. It means that our prediction differ so much from reality that it is pretty impossible to meet the commitments arranged with the Electricity Market the day before.
Case 4: Overproduction + Storage Full
Let’s go directly to the Simulation:
As it can be seen, it has been rounded with a blue circle (third plot: control action U) the moment when the controller instantaneously stop the loading process because the upper limit of our Storage is reached. As a result, the control action takes a null value. Rounded in green, it can be seen the unavoidable power surplus that is supplied to the Electricity Grid (first plot). In red (second plot) we find that the upper bound (restriction of 200 MWh of maximum storage) is severely respected by our Predictive Control Algorithm.
It is up to the Plant Operator whether or not to supply that power surplus. One option could consist on disconnecting some Solar PV Modules in order to low the Energy Production. Other solution could be to drive that energy excess to other applications.
Case 5: Lack of Generation and Insufficient Storage
Let’s finish with the opposite situation to the one shown in Case 4. What happens if we fail to predict the lack of Sun and, to make the matter worse, we do not have enough Energy stored in our batteries?
The simulation would be as follows:
Rounded in black in the second plot we have selected two different moments when the Stored Energy is very close to zero. During the operation day, our Predictive Controller has been unloading energy from the Storage but, it comes to points when it cannot be discharged anymore. As a result, we will fail to meet our hourly commitments at some time slots, as it is shown in the following graph.
The idea that could well summarise the exposed results is as follows: the proposed strategy is able to handle different kind of adverse non-predicted situations in terms of weather, although there are still physical limits that could make some cases impossible to “solve”. However, our Predictive Algorithm subject to constraints has proven good performances in the majority of the simulations.