Dynamic operation

Based on the previous chapter, the operation of the model is satisfactory and the dynamic performance can be studied. The control system developed in Chapter 4.4 may be the most detailed part of the model, and therefore contains multiple features that could be highlighted. The simplest task is to introduce step changes to the controllers responsible for power consumption and generation. In Figure 47 and Figure 48, 10% and 25% step changes are respectively introduced for the compressor

50

train power consumption set point value. Before the step change at t = 5 minutes, the controllers are given sufficient time to stabilize, so that the fluctuations in each of the compressor inlet temperature are only caused by the control operation. From the figures, one can observe that the power generation follows the set point very closely.

Marginal overshoot spikes can be noticed by closer inspection from both the figures, but the magnitude is practically negligible.

Figure 47. Evaluation of compressor train control performance, 10% step change in power consumption set point introduced at t = 5 minutes.

Although PID controllers are not commonly used in temperature control, the complications of variable guide vanes with controller deadbands described earlier led to the selection (Hannu Mikkonen 2015, personal communication). The PI control typically used is inherently more fluctuating, and even small deadband values of 0.5ºC often led to a situation in which the system took excessive time to stabilise.

With PID control, the fluctuations are minimised and the temperatures stabilise nearly perfectly in 10 minutes regardless of the magnitude of the step change. This is partly caused by the thermal inertia of the heat exchangers – when the air mass flow

18.5

rate is decreased, the relatively large amount of accumulated heat enables relatively low variation in thermal oil mass flow rate. With both step changes, the greatest variation in temperature is in last two stages, which are tuned with considerably more aggressive parameters than the first two stages. For example, the last stage is

Figure 48. Evaluation of compressor train control performance, 25% step change in power consumption set point introduced at t = 5 minutes.

For expanders, step changes of the same magnitude are introduced and the resulting control performance is shown in Figure 49 and Figure 50. As described earlier, the power is regulated by a cascade scheme of PI controllers. The issue with PI controllers in general is the tendency to overshoot, particularly when tuned using the Ziegler-Nichols method, which can be noticed from the both the figures. As the controller reaches the set point value in power generation, the throttle pressure continues to decrease, forming a sharp spike before stabilising.

12.5

When the throttle valve rapidly changes its position slightly after t = 5 minutes, particularly the first stage temperature decreases relatively quickly, forming a vertical drop in the curve. Similarly, when the valve reaches the minimum point in pressure, the temperatures increase with a similar rate of change. This phenomenon does not take place in the line before the throttle valve, suggesting that it can be considered to be due to throttling instead of being a numerical error. Moreover, the total friction coefficient of the valve shows a sharp spike when the position of the valve is altered quickly, further supporting the idea. As the throttle pressure continues to decrease, each of the temperatures experience behaviour which is best described as sliding. Although the temperature controllers react to this disturbance instantly, the inertia also noticed with the compressor train does not allow the temperature to be regulated immediately. This can be observed as slightly curved increase in the temperature before the second spike.

80

Figure 49. Evaluation of expander train control performance, 10% step change in power generation set point introduced at t = 5 minutes.

Figure 50. Evaluation of expander train control performance, 25% step change in power generation set point introduced at t = 5 minutes.

The first expansion stage is subject to greater disturbances due to its physical location after the throttle valve, which is controlled with a driving time of one second and cascade control with considerably short integration times. Despite of this, the fluctuations are within the range of 2ºC even with the greater step change. The temperature control is carried out with a single output PID controller unlike in the compressor train. Cascade PI control scheme based on thermal oil mass flow rate and expander inlet temperature was evaluated as well, but due to its stronger oscillations and the consequent effect on the fast throttle valve it was not selected. Particularly the temperature controllers of the first two stages show good performance, as the set point value is maintained after the second wave without using any deadband. The controller of the third stage was considered more problematic and a deadband of 0.1ºC was applied to stabilise the operation. The controller is able to find the deadband range after the first wave, which can be perceived as the curved increment in the temperature.

The greatest challenge to the control system is the grid operation, which combines the requirements to regulate the power consumption and generation according the

80

demand, and to make decisions to operate the system at the correct time.

Simultaneously, the turbomachinery inlet temperatures have to be controlled at all times, and the surge margin during the compressor start-ups. In Figure 51 and Figure 52, the model is operated for six hours in dynamic conditions in accordance to the input data introduced in Chapter 4.3.5. The former figure shows the power charged to or from the grid depending on the excess or deficiency in wind generation, while the latter figure illustrates the state of compressed air storage and TES.

Figure 51. Power consumption and generation depending on the excess or deficiency in wind generation during the dynamic operation.

Figure 52. Pressure of compressed air storage and fill percentage of TES tanks during the dynamic operation.

As mentioned, the wind data is scaled in order to evaluate each part of the control system and the transitions between the operation modes. The following operation is primarily expected from the model during each of the temporal segments:

- Hour 1: nominal charging - Hour 2: no operation - Hour 3: nominal charging - Hour 4: part-load charging - Hour 5: part-load discharging - Hour 6: discharging

The results shown above indicate that the expected operation is to large extent fulfilled. For example, during the last hour of the operation Figure 52 shows the compressed air storage pressure steadily decreasing, while the thermal oil is transferred from the hot TES tank to the cold TES tank at decreased temperature level. The flat plateaus in Figure 52 indicate that the system is not active, whereas the predominant flatness of the area curve in Figure 51 suggests that the system is to large extent operated with nominal load rate. The capability for accurate load following is particularly highlighted at around t = 2 h, where the high variability of wind leads to rapid changes in the wind farm output, to which the system is able to respond. Furthermore, even though the operation is only scheduled by using 10-minute-ahead wind forecasts, the false signals for start-ups and shutdowns are largely avoided. This is proven during t = 1 … 2 h; although the system could potentially fulfil the deficiency in wind generation, the predictive boundary conditions prevent the system from starting as sufficient time for consecutive operation is not expected.

On the other hand, exceptions of this are visible e.g. slightly before t = 3 h, where the system is needlessly shut down and started up multiple times consecutively. The problem, however, could be easily solved with higher degree of interlock mechanisms, for example through inclusion of longer-term forecasts.