1. Introduction
Historically in Germany, electric energy was supplied by large conventional power plants in the transmission grid to consumers mainly connected to the distribution grid. Distribution grids were therefore dimensioned to ensure power supply in extreme load cases. In this context, the clear division of transmission and distribution operation and planning is reasonable.
In the year 2000, the German Renewable Energy Act was introduced in order to reduce greenhouse gas emissions, leading to a strong expansion of renewable energy (RE) systems, which is expected to be continued in the future. By 2050 at least 80% of Germany’s electricity demand is supposed to be supplied by RE production [
1]. Due to their relatively small nominal power these power plants are mainly connected to the distribution grid (e.g., [
2,
3]). This substantial transition of energy supply structure leads to more diverse power flow characteristics (e.g., [
4,
5,
6]). In simple terms, unidirectional flow patterns are shifted to bidirectional ones. Distribution grids not only distribute but also supply power changing their behavior with respect to the mostly weather-dependent renewable power production. The increase of complex interdependencies between the different voltage levels calls for research on integrative transmission and distribution grid planning (see also [
6]).
Furthermore, the described energy transition most likely leads to substantial grid and storage expansion needs in the German power system (e.g., [
2,
3,
7,
8,
9,
10]). For society it is of interest to find economically efficient solutions. In this spirit, a combined economic optimization of grid and storage expansion considering technical constraints seems to be a worthwhile task.
On the one hand, the need for combined optimization of grid and storage expansion and on the other hand the spatially detailed integrative consideration of all voltage levels motivate this work and lead to the following research questions:
How is it possible to integrate transmission and distribution grid planning?
Is it (systematically) cost-efficient to align distribution grid planning with the overall systematic optimization of power plant dispatch, grid and storage expansion?
This work explicitly uses and creates open data and open source tools. The underlying paradigm is motivated by multiple considerations. Due to the lack of available grid data, open data methods enable and enhance independent and diversified research. Additionally, the good scientific practice aims to create reproducible methods and results. Last but not least, the energy transition displays a fundamental societal effort which may be better addressed in a transparent way to raise public acceptance for scientifically reasonable changes. Recently these aspects have been addressed more often in energy research gaining greater relevance (e.g., [
11,
12,
13,
14,
15]).
The following
Section 2 provides an overview of the state of the art of transmission and distribution grid planning. Here, we primarily focus on the research and best-practice principles which are conducted and applied by transmission and distribution grid operators. Thereafter, a novel approach of integrated transmission and distribution grid planning is presented (
Section 3). First, a brief introduction to the underlying grid model is provided. Second, the methods used to optimize grid and storage expansion on the extra high (EHV) and high voltage (HV) level are described. Third, the interface to the medium (MV) and low voltage (LV) level is characterized. Lastly, the grid expansion, storage and curtailment distribution methods on these lower distribution grid levels are specified. Based on this we present, discuss and critically address our results for a mid-term and a long-term future scenario (
Section 4 and
Section 5) before we conclude our findings in the last section.
5. Discussion and Critical Appraisal
In the following we discuss and verify our results by comparing them to other studies and articles. We separately consider state of the art transmission and distribution grid planning for Germany as grid level integrated and at the same time spatially high-resoluted approaches have not existed before. Due to our innovative approach, quantitative comparisons have to be conducted carefully. Therefore, the results are indicatively interpreted and the most important methodical differences analyzed. Moreover, we critically address sensitive methods and assumptions.
When comparing the results for grid expansion on the EHV and HV level in the
NEP 2035 scenario with the results of the official
NEP 2035 scenario [
7] the grid expansion stays remarkably below the official plans to invest 20 to 36 bn EUR into new grid infrastructure. This has various reasons. In our model no restrictions on a minimal RE share or maximal RE curtailment are assumed. In contrast to our results the official NEP has a higher RE share of 67% and only allows that each RE power plant is curtailed by 3% of its yearly potential energy production. In contrast, our in this context unconstrained optimization allows an overall curtailment of 19% of the potential wind and solar energy production. The biggest curtailment of a single RE aggregate adds up to 88% throughout the year. In the given setting this is a macro-economically efficient solution but would not fit the current market design. Another difference displays the electricity trade balance. The official NEP simulates a net exporting German system whereas our results show an importing behavior. In our co-optimized approach this also leads to less grid expansion within Germany. Finally, instead of a determination of discrete grid expansion measures considering realistic component sizes, as realized by the NEP, the results were continuously optimized and not discretized afterwards. This surely leads to an underestimation of grid expansion costs. Moreover, we analyzed cost reducing effects being motivated by a different technology usage. First, the line lengths of the built capacities play an important role. In the optimization short lines are therefore preferably expanded. Second, we solely consider AC grid expansion on the existing routes expecting new overhead lines. In contrast, the NEP focuses on investing into underground DC technology, which according to literature is specifically (per MVA*km) 17 times more expensive (including converter costs and the average length of the officially planned DC link projects (
370 km) considering cost assumptions of
Table 1). Aside from the grid expansion, the lacking need of storage expansion in this context stays in line with the NEP and the discussed findings in [
43].
Concerning the
eGo 100 scenario, the grid expansion of 6.8% of today’s grid capacity is significantly higher compared to the
NEP 2035 scenario but stays below the official NEP results. The general reasons for calculating less investment costs stay the same. The storage expansion of 13.7 GW is similar to the results in [
43] for an RE share of 67% but without considering any grid expansion. Compared to other 100% RE scenarios the modelled storage expansion is rather low. For example in [
75] the results for 2050 range between 13–39 GW. In a meta study [
76] a range of 20–94 GW with respect to several studies was shown. Ref. [
77] states, grid expansion may substantially reduce the need for storage expansion. The results of the combined grid and storage optimization emphasize these findings. The significant curtailment of fluctuating RE indicates remaining potential for additional storage investment. Nevertheless, the high reliance on exchanging electricity with the neighboring countries and the corresponding net import make storage investments less attractive. In this context the gas fired power plants in neighboring countries influence the results. Constraining the system to less possible net import would probably increase domestic grid and storage investments. Moreover, another driver for a lack of storage investment represents the significant share of flexible RE resp. biomass and hydro supplying balancing power. In the context of a rising relevance of coupling the electricity sector with other sectors such as heat and mobility, this flexibility might be needed for other purposes. For example in [
10] biomass is not considered for any electricity production reserving it for other hard-to-defossilize sectors (e.g., shipping, aviation). In future research the effects of sector coupling and other variations and sensitivities shall be analyzed.
The results in the MV/LV level revealed that the applied flexibility options can be cost-effective. Nevertheless, over all MV grids the cost reduction effect was marginal. This apparent controversy shall be discussed in more detail. As described, distributed storage units for possible allocation were only available in a few MV grids. The majority of the storage investment is economically most feasible as large-scale long-term storage. These are not assumed to be possibly allocated on the MV level. Concerning curtailment the optimization results on EHV/HV level also limit the potential for savings on MV/LV level. Another reason for the small overall effect leads back to parametrization. For the sake of runtime reduction only the HLF and RPF were considered out of the time series to determine grid expansion needs. It is possible that the HLF and RPF snapshots differ between the
Flex and the
Reference scenario and that this simplified approach may disregard snapshots which are relevant for the grid dimensioning. Consequently, in some cases the grid expansion costs are even higher in the
Flex than in the
Reference scenario. Nevertheless, the tendency is such that, the bigger the reduction of the maximum RPF, the more investment cost savings can be achieved. In addition to curtailment and storage integration, many studies have investigated other flexibility options such as reactive power management, on load tap changers, or demand response that can as well significantly lower grid expansion needs in distribution grids. Besides a fixed cos
(P) of 0.9 (inductive) applied here that [
78] showed to potentially reduce grid expansion costs, these measures were not included in this study as focus is on analyzing how measures that have proven to be cost-optimal in the overall systematic optimization can be utilized in the underlying voltage levels to reduce grid expansion costs there as well.
In contrast to the low difference between the
Flex and the
Reference scenario, the savings compared to worst-case scenarios are enormous. Since the worst-case scenarios try to reflect the state of the art distribution grid planning (see
Section 2.2) we can compare these results with literature in order to validate them to some extent. In Germany two major distribution grid studies have been conducted by dena [
3] and BMWi [
79], analyzing scenarios that are in their share of RE comparable to our
NEP 2035 scenario. Thus, worst-case grid expansion costs calculated in our
NEP 2035 - Worst-Case-600 scenario for MV and LV levels of 15.2 bn € lie in between costs calculated by dena for the MV and LV levels of 11.4 bn € (
NEP B 2012 scenario, target year 2032) and costs calculated by BMWi of 17.3 bn € (
NEP scenario, target year 2032). It has to be noted that a comparison with these scenarios can only serve as an indication as some key parameters such as newly installed solar and wind capacity differ significantly. dena and BMWi do not investigate 100% RE scenarios wherefore a direct validity check for the
eGo 100 - Worst-Case-600 cannot be made. However, dena analyzed a scenario with a RE share of 82% for which they determined grid expansion costs on the MV and LV levels of 16.2 bn €. Compared to this costs of the
eGo 100 - Worst-Case-600 of 19.1 bn € can be considered reasonable.
Consequently, the modelled substantial savings of our
Flex towards these state of the art scenarios indicate a significant potential for basing grid planning on spatially differentiated time series instead of using general simultaneity factors. However, this outcome has to be critically addressed. In order to correct (in terms of potential yearly energy production) the overestimation of the coastdat-2 weather data, correction factors were introduced [
80]. Due to a linear down scaling, the maximal potential weather-dependent power output is reduced accordingly. Since the maximum RPF is highly significant for the grid expansion evaluation in the MV and LV levels, especially the wind onshore and solar reduction factor of 0.6 and 0.8 most probably lead to an underestimation of the modelled grid expansion costs within the MV and LV level.
With the help of the MV clustering the results for the entire German MV/LV level could be derived from the calculation of representative MV grids. As observed in
Figure 5 for the worst-case approaches, a greater number of considered grid representatives lead to more accurate approximations. Assuming this behavior also for our more sophisticated approach in the
Flex-scenario we chose a setting with 600 MV grids. This setting required two weeks of parallelized calculation on 31 threads needing up to 250 GB random access memory. A setting of 20 cluster grids can be calculated in parallel in less than 24 h. This less extensive setting revealed 8% more grid expansion costs for the MV/LV level in the
eGo 100-Flex-scenario compared to the calculation with 600 MV grids. In the worst-case evaluation the calculation with 20 grids overestimated the full calculation by 19%, whereas the calculation with 600 grids only showed an overestimation of 1%. Thus, in the
Flex scenarios the overestimation tendency seems to stay alike but less dominant.
In this work, complexity reduction plays an important role. The spatial complexity is reduced not only on the MV level but also on the EHV/HV level by the usage of k-means algorithms. This method only finds local optima and is thereby sensitive to the initial random guess of cluster centroids. We minimized this effect by choosing many iterations with different initial settings and small inertia tolerances in order to get more robust results. Furthermore, the computing of Euclidean distances makes the MV clustering sensitive to outliers, i.e., MV grids with extraordinary attributes. To avoid this, a utilization of a k-medoid algorithm can be reasonable [
81]. However, [
72] showed that the k-means algorithm works well for the estimation of grid expansion costs. Undoubtedly the complexity reduction may induce inaccuracies. The usage within the EHV/HV level leads to neglected local grid restrictions within one cluster. In [
82] the disregard of the intra-zonal grid topology was shown to be rather high. In contrast, [
9] analyzed that the increasing behavior of grid expansion costs flattens out when exponentially increasing the number of clusters (up to 362) and concludes that the clustering captures crucial transmission routes well even when considering a smaller number of clusters. Hence, with 300 grid nodes this effect should be acceptably low. Moreover, as shown in [
6] the consideration of the HV level includes significantly more grid restrictions compared to a sole focus on the EHV level.
In general, the modelling relies on manifold assumptions and methods. The LOPF as a central modelling approach for the top-level optimization has been verified in many previous works [
6,
9,
10,
43,
66,
83,
84]. The developed interface to the MV level uses continuous consistency and plausibility tests to ensure code integrity. Moreover, the non-linear AC power flow has been effectively used for many decades [
66,
85] and the grid expansion heuristic used on MV/LV level is as well state of the art [
3,
31]. A sensitivity analysis on various parameters (e.g., temporal and spatial clustering settings, storage costs, electricity consumption) considering all voltage levels, as well as simplified verification cases for the curtailment and storage integration heuristics showed plausible results [
86]. Furthermore, the advantages of focusing on open data such as OSM come along with the problem of difficult validation. Previous works such as [
14] (EHV/HV data model), [
42] (MV/LV data model) and moreover [
6,
43] have critically discussed these possible inaccuracies of the underlying data basis. This problem should be born in mind and remains to be further addressed.
6. Conclusions
This work presents a novel approach to optimize power plant dispatch, grid and storage expansion considering all voltage levels of the German public power grid. The HV level is fully integrated into the top-level LOPF jointly optimizing dispatch, grid and storage expansion while meeting the linearized network equations. Hence, the traditional distinction between distribution and transmission grid is partly dissolved. Afterwards, to obtain the reactive and lossy power flow behavior and to ensure a physically feasible power system design a non-linear PF is performed. Consequently, we present a unidirectional interface towards the MV level, which enables a consistent optimal distribution of new storage units and curtailment measures being in line with the EHV/HV optimization results. In order to derive results for whole Germany while keeping computational effort manageable, complexity reduction methods were developed on the EHV/HV as well as for the MV/LV level.
In a mid-term scenario for the year 2035, storage expansion is not an economically feasible option. For a 100% renewable power system in Germany 14 GW storage and 114 GVA grid expansion are co-optimal considering an EHV and HV grid model. Taking into account all voltage levels, total grid expansion investment of 18.6 bn EUR and storage expansion expenses of 11.1 bn EUR are cost-efficient.
Due to the unidirectional top-down approach the cost-efficient usage of flexibility options i.e., storage expansion and curtailment is limited by the top-level results. Thus, the results for the MV and LV level are twofold. In the eGo 100 scenario the battery storage distribution method leads to a considerably high average reduction of grid expansion costs by 22% concerning the particular MV grids where the decentral battery storage units at the HV/MV connections are also economically feasible for the overall system. Only for 0.9% of all MV grids in Germany this prerequisite was given. Accordingly, the overall investment cost saving was only marginal. In this context, the optimal usage of curtailment did also not reduce the overall MV grid expansion costs significantly although in some grids the cost savings are reaching 30%. Reasons for this effect have been discussed and need to be further analysed.
Comparing the results to grid expansion costs derived by a state of the art worst-case approach, substantial savings depending on the scenario of about 50% were calculated. It can be concluded that the usage of realistic, spatially differentiated time series instead of using general simultaneity factors leads to significant saving potential for distribution grid planning. Nevertheless, this outcome is biased by a linear downscaling of the potential weather-dependent generation time series, minimizing potential and relevant worst cases. As a future research it will be a task to find a more sophisticated correction method which would reflect the weather behavior more accurately. Further research will also be directed toward including the expected electrification of the heat and mobility sectors in the future scenarios that might significantly increase the load in the distribution grids. This could further increase the relevance of an integrated grid planning in order to leverage the flexibility potential these new loads could provide and keep resulting grid expansion needs at a minimum.