**4. Results**

The results of the predictive simulation-based optimization are compared in five different scenarios. The first scenario is the reference case, which is based on the energy monitoring data of the plastic processing company. In the reference case, the company is located in Kassel, Germany, and has a fixed price for electricity. In the second scenario, the company is also located in Kassel, and the predictive simulation-based optimization is applied to it. To quantify the influence of the prediction horizon, it is varied between 6 and 48 h. To quantify the influence of the ambient temperature on the predictive simulation-based optimization, scenario 3 is identical to scenario 2, except the company's location is changed from Kassel to Madrid, Spain, and the prediction horizon is set to 48 h. Further, scenario 4 shows the influence of an increased ambient temperature of Kassel by 5 K. Scenario 5 is similar to the second one, except the company purchases its electricity on the EPEX SPOT market and has a variable price. The focus of this scenario is to analyze the influence that a variable electricity price has on the operating schedule of the different chillers. In scenario 6, the company is in Madrid and purchases its electricity from the EPEX SPOT market. Table 2 shows an overview of the different scenarios and the varied parameters of the simulations. The cooling demand for the molding cooling circuit is in all scenarios identically, since hygiene standard forbids any window opening, and therefore the production hall is air-conditioned to a required temperature of 21 ◦C.

**Table 2.** Overview of simulated scenarios and parameters.


#### *4.1. Prediction Horizon*

Figure 5 shows the different operating hours of the three chillers for a prediction horizon *n* between 6 and 48 h. The bottom bar represents the operating hours of the free cooling chiller. The ascending bars show operating hours of the first compression chiller for each stage and the top bar refers to the second compression chiller. Stage 4 of chiller one and stages 2–4 of chiller two are not used in any of the four cases.

**Figure 5.** Operating hours of chillers in different stages for varying prediction horizons.

It illustrates that the increase in hours of the prediction horizon leads to a rise of the operating hours of the free cooling chiller and increases the operating hours of the two compressions chillers in stage one. The influence on the operating schedule is discernible.

Due to the better efficiency of the free cooling chiller compared to the compression chillers, the electrical energy consumption decreases proportionally to its operating hours. In comparison, the reference case with standard control is also visualized. Figure 6 illustrates the total energy consumption for the four cases, with differing prediction horizons in scenario 2, the reference scenario, as well as the average EER for the year. The results show that the optimization reduces electrical energy consumption by around 38% (N6–N24). A prediction horizon of *n* = 48 decreases the electrical energy demand by another 5% because of the better charging and discharging strategy of the sprinkler tank.

**Figure 6.** Energy savings and energy efficiency ratio (EER) comparison for varying prediction horizons.

The difference in energy efficiency and the different operating hours of the free cooling chiller and the stages of the compression chillers result from the charging and discharging strategies based on the different prediction horizons. Figure 7 shows the charge level of the storage tank for the low supply temperature at 14 ◦C (black line) depending on the time of the year in hours for the four different prediction horizons. The charge level of the high temperature water of 17 ◦C is the mirror image of it. A prediction horizon of 6 or 12 h results in a minor usage of the energy storage potential of the tank. In the case with a prediction horizon of 24 h, the storage is charged and discharged over 50% of its full capacity in the first and fourth quarter of the year. The full capacity of the storage is used in the case with a prediction horizon of 48 h.

**Figure 7.** Charge level of the storage tank for varying prediction horizons.

The complete charging and discharging of the tank increase the running hours of the free cooling chillers (Figure 5). With a long prediction horizon, the tank is pre-charged when ambient temperatures fall below 11 ◦C and discharged in times of high ambient temperatures, minimizing the load on the compression chillers and even placing them in standby mode. This strategy leads to the extra 5% saving potential by extending the prediction horizon to 48 hours.

#### *4.2. Influence of Ambient Temperature*

The efficiency of the free cooling chiller and the air-cooled compression chillers depend on the ambient temperature. Lower ambient temperatures allow lower condensing temperatures, resulting in better efficiency. If the ambient temperature is higher than 11 ◦C, the free cooling chiller does not provide the required cold-water supply temperature. Both effects have an influence on the energy-saving potential of the predictive simulation-based optimization. Hence, in scenario 3, the plastic processing company is moved to Madrid, Spain, to have a different ambient temperature profile. In scenario 4, the German temperature profile is lifted by 5 K. Figure 8 shows the total energy consumption and EER for one year in comparison to the reference case and scenario 2. All optimizations are based on a prediction horizon of 48 h, since the previous analyses in Section 4.1 conclude that this is the best horizon to maximize energy savings.

The change in the ambient temperature reduces the energy-saving potential for Madrid to 23% and, for scenario 4, to 19% compared to the reference case. Because of reduced running hours of the free cooling chiller in scenarios 3 and 4, the energy efficiency is significantly lower compared to scenario 2. However, the analysis of the different ambient temperatures also shows that the

predictive simulation-based optimization together with the sprinkler tank reduces the electrical energy consumption in warmer regions. Although the energy-saving potential is reduced for warmer ambient temperatures, there is still a significant reduction compared to the reference case.

**Figure 8.** Energy savings and EER comparison for different ambient temperatures.

#### *4.3. Variable Electricity Prices*

In the previous scenarios, the electricity price for the plastic processing company is fixed to 16 euro cent/kWh. In scenarios 5 and 6, the electricity procurement is via the day-ahead exchange market in Germany. All taxes, costs, and the German renewable energy levy are included in the model. To benefit from purchasing electricity with a variable price, it is required to shift loads. The sprinkler tank enables the cold utility system to store cold-water in times of little or no demand. This allows the optimization algorithm to start chillers in times of low electricity prices and stop the machines in times of high prices. This increases the optimization options. Figure 9 shows the results of the simulation for the variable cost option in comparison to the fixed price and the reference case. On the left *y* axis, the thermal and electrical energy consumptions are visualized, and the costs are shown on the right *y* axis.

**Figure 9.** Comparison of variable and fixed price optimization.

Under scenarios 5 and 6, the electrical energy consumption for Kassel increases by 7% and, for Madrid, by 0.4%, but the total energy costs for the cooling systems are reduced. For the plastic

processing company in Kassel, the electricity procurement via the spot market reduces the costs by around 9.3%. For the site in Madrid, the costs reduction is about 12%. The increased electrical energy consumption is due to the reduction of the operating hours of the free cooler and the simultaneous increase in the use of the compression chiller. This leads to a decreased overall efficiency of the system. However, the lower efficiency is compensated by the lower prices for the electrical energy on the spot market in comparison to a fixed electricity price. Figure 10 shows the comparison of the charge level of the sprinkler tanks for fixed and variable price scenarios.

**Figure 10.** (**a**) Charge level of storage tank for a fixed electricity price; (**b**) charge level of storage tank for variable price scenarios.

In Figure 10b, the electricity price is shown on the right *y* axis. In comparison to the fixed price scenario, the storage tank is never fully charged in the variable price scenario. Although both prediction horizons are equal, the results of the optimization differ for minimizing the costs. The different charging and discharging strategies are the reason for different electrical energy consumption and the different total energy costs.

#### *4.4. Results Summary*

Table 3 summarizes the results for the comparison of the different prediction horizons, the influence of the ambient temperature, and the effect of purchasing electricity for a variable price. The savings in percentage for energy and costs use the basis of the reference case.


**Table 3.** Overview of optimization results.

#### **5. Conclusions and Outlook**

The study developed a predictive simulation-based optimization of a cooling system with continuous self-learning performance curve models that saves electrical energy by optimal charging and discharging of the sprinkler tank. The optimization reduced electrical energy costs via electricity procurement on the EPEX SPOT market. Results for the case study concluded that the control strategy of the optimization together with the installation of cold-water storage tanks with a significant volume and a free cooling chiller could save over 43% of the electrical energy in comparison to the reference case. To utilize the full potential of the storage tank, a prediction horizon of 48 h was necessary. In comparison, a prediction horizon of 6 h saved up to 38% of electricity use. The difference in the prediction horizon had minimal impact on the computation time. A longer prediction horizon increased the energy savings, but the available thermal storage capacity limits it. In further analyses, the influence of the modeling error and weather forecast changes on the optimal operational strategies as well as the potential for exergy and entropy considerations may be investigated.

The study further investigated the effects of ambient temperature due to different plant locations as well as a variable electricity spot price. The ambient temperature profile had a significant influence on the energy-saving potential. The higher the average ambient temperature, the lower the energy saving—however, the optimization strategy still saved around 20% in warmer regions. Electricity procurement via the EPEX SPOT market led to a slight increase in electrical energy consumption but saved 9.3% to 12% of the energy costs. For the analyzed plastic processing company, the differences in the electricity price were low but still influenced the cooling utility system. If the company should economically support the electricity grid, the price differences need to be higher or include extra incentives. For example, a dynamic tax on electricity based on the instantaneous share of renewable electricity generation presents a chance to increase the differences in the variable prices and make demand-side managemen<sup>t</sup> more attractive for medium-sized companies.

**Author Contributions:** Responsible for the conceptualization, R.-H.P. and F.S.; methodology, R.-H.P. and H.M.; validation, R.-H.P.; investigation, R.-H.P. and H.D.; data curation, H.D.; writing—original draft preparation, R.-H.P.; writing—review and editing, F.S., H.M. and T.G.W.; visualization, R.-H.P.; supervision, T.G.W.; funding acquisition, R.-H.P.

**Funding:** This research received no external funding.

**Acknowledgments:** The authors gratefully acknowledge the financial support of the Rud. Otto-Meyer-Umwelt-Stiftung and the support of the EU project "Sustainable Process Integration Laboratory – SPIL", project No. CZ.02.1.01/0.0/0.0/15\_003/0000456 funded by EU "CZ Operational Programme Research and Development, Education", Priority 1: Strengthening capacity for quality research.

**Conflicts of Interest:** The authors declare no conflict of interest.
