Prepare data for pathway optimisation with sector coupling

[nworbmot]

Heating

• Today's split of heating technologies in each country: district heating / gas boiler / oil boiler (JRC IDEES in "residential" and "tertiary")
• Approximations of age distribution of heating infrastructure (assumption e.g. 25% 5 years old, 25% 10 years old, etc.)
• Capital costs and fuel costs over time

  Transport

• Number of vehicles in each country (JRC IDEES)
• Distribution of their ages, or make assumption (e.g. 25% 5 years old, etc.), could be different in different countries
• EITHER cost prediction for EVs over time (e.g. from Bloomberg NEF or Fraunhofer IEE or EnergyPLAN) OR exogenous share of EV over time

  Industry

• Costs of different production pathways, e.g. blast furnace versus DRI + electric furnace, to determine substitution pathway (Christian Breyer has researched some of these costs) OR exogenous pathway

Other studies

Fraunhofer IEE (formerly IWES) has cost assumptions over time for vehicles, heating, etc., as does Palzer PhD thesis, see e.g. references in Synergies of Sector Coupling paper, e.g. http://www.energiesystemtechnik.iwes.fraunhofer.de/de/projekte/suche/laufende/interaktion_strom_waerme_verkehr.html has diesel versus petrol versus EV for DE until 2050, see e.g. Figure 0-9, costs in Table 10-35 (beware battery costs too high since study is old).

[nworbmot]

This is now done for electricity and building heating in scripts/add_existing_baseyear.py. Still to do:

• [ ] Assumptions for a split rooftop-utility for the existing solar
• [ ] Assumptions for the split between AC- and DC-connected offshore for existing offshore (only really Germany has DC-connected offshore)
• [x] Use updated decommissioning dates from powerplantmatching that include planned nuclear and coal shutdowns
• [ ] Use retrofitting date for power plants to extend lifetime (@martavp suggests half of lifetime after retrofitting)
• [ ] Use real distributions of existing wind and solar rather than a capacity factor heuristic
• [ ] The share of heat pumps in meeting heat demand in 2020 in residential and services urban decentral seems high (50%) - not sure whether this is result of optimization or a problem in underlying data on existing heating infrastructure
• [ ] Could use energy split for heating in 2015 in JRC-IDEES as a guideline for existing heating capacities and for district heating shares. For capacity could just assume it meets peak with 10% buffer.
• [ ] Since buildings are only supplied with a single technology today (gas/oil/electricity etc), for existing assets, it probably makes sense for them to supply with a fixed profile proportional to the heat demand with p_max_pu = p_min_pu, since we want to avoid inadvertent mixing of heat sources. Gurobi will see these supply variables are constant and remove them. For greenfield assets, we can mix sources with hybrid heat pumps and combinations of heat pumps and resistive heating. (Is mix for each build year fixed?).
• [ ] A similar approach of constant supply can be used for existing transport and industry supply? And for exogenous steel fuel switching?
• [ ] Squashing assets with p_nom_extendable=False and the same attributes (including time-dependent attributes like p_max_pu), except p_nom and build_year, into a single asset to reduce the number of variables. We'd need a script to squash assets before solving, and unsquash them afterwards. I have a code sketch, but don't want to implement it right now.
• [ ] Have time-dependent configuration attributes for investment years 2020, 2030, 2040, 2050, e.g. exogenous district heating share, EV share, share of EVs doing V2G or DSM, steel with DRI+EAF, etc.. This could be done in same file as CO2 budget for each investment year.
• [x] Add a rule to create the CO2 constraint path based on a CO2 budget (e.g using exponential decay, S-curve). @martavp has a draft script for this.
• [ ] Add the calculation of cumulative cost for the transition path to the rule make_sumary. The sum should use a social discount rate to be included in the config file.

[nworbmot]

For the squashing of assets with the same attributes to reduce the number of optimization variables, my suggestion would be:

• Prepare the prenetwork-brownfield as normal
• Have a script that aggregates similar components to representative components before saving the network to prenetwork-brownfield-aggregated
• Solve prenetwork-brownfield-aggregated to postnetwork-aggregated
• Have a script that disaggregates postnetwork-aggregated to postnetwork (which has the same components as prenetwork-brownfield)

For the aggregation, e.g. all generators with the same bus, marginal cost, efficiency, p_min_pu and p_max_pu would be aggregated to a single representative. The existing p_nom's would be added up. If one of the generators is p_nom_extendable, then the representative is also p_nom_extendable.

For the disaggregation, all the new capacity goes to the generator with p_nom_extendable (if more than one is p_nom_extendable, it can be split, but in our use cases, at most one should be p_nom_extendable). The dispatch n.generators_t.p is split to the generators proportional to their p_nom.

Here's code for identifying the mapping from generators to representative aggregated generators:

#attributes that should match
attributes = {"Link" : pd.Index(["marginal_cost","p_max_pu","p_min_pu"]).append(n.links.columns[n.links.columns.str.contains("efficiency")]).append(n.links.columns[n.links.columns.str.contains("bus")]),
              "Generator" : pd.Index(["marginal_cost","bus","p_max_pu","p_min_pu","efficiency"])}

for c in n.iterate_components(attributes.keys()):

    #store the mapping to representatives here
    mapping[c.name] = pd.Series()
    selection = c.df[attributes[c.name]]

    #to compare time-dependent, just check the average
    for attribute in attributes[c.name]:
        if attribute in c.pnl and not c.pnl[attribute].empty:
            selection.loc[c.pnl[attribute].columns,attribute + "_t"] = c.pnl[attribute].mean()

    for i in selection.index:
        found = False
        for j in mapping[c.name].index:
            if i[:-4] != j[:-4]:
                continue
            if selection.loc[i].equals(selection.loc[j]):
                print(i,"is same as",j)
                found = True
                break
        if not found:
            mapping[c.name][i] = i
        else:
            mapping[c.name][i] = j

The new representative components are given by mapping[c.name].unique(). All others should be dropped and p_nom added to the representative component. Check for p_nom_extendable.