Skip to content

Model limitations

FlexTool is a linear (or mixed-integer) optimisation model of an energy system. Like every model, it leaves some things out and simplifies others. This page lists the limitations that matter most in planning studies, with the usual way around them. Read it before you trust a result that depends on one of these features.

Emissions

  • CO2 is the only emission. Each commodity has one constant co2_content [tCO2/MWh]; other pollutants are not tracked.

    The emissions of a group come from the fuel its units draw from the group's commodity nodes. A cap or a price for a sector therefore needs separate fuel nodes for that sector. The emissions of a unit with several products (combined heat and power, cogeneration with desalination) are not split between the products.

  • A CO2 limit has no slack. If the system cannot meet co2_max_period or co2_max_total, the model leaves demand unserved at the node's penalty_up instead (or is infeasible when the node's penalty_method is off).

    Check the slack results of a run with a tight limit; see How to compare emission trajectories.

Thermal units and unit commitment

  • Linear commitment is an approximation. With startup_method linear, a part of a unit can be online: with min_load 0.5, a 500 MW unit can run at 25 MW with 50 MW (10 %) of it online.

    Minimum load and start-up costs apply only to the part that is online, so a large unit looks more flexible and cheaper to cycle than it is. binary commits whole units (a mixed-integer problem): it is exact, but slower. virtual_unitsize splits a fleet into smaller units to keep a binary model fast.

  • No start-up fuel. A start-up costs only startup_cost [CUR/MW]: it uses no fuel or heat, and there is no shut-down cost and no hot, warm or cold start.

  • Fuel use is linear in the output. With conversion_method min_load_efficiency, the fuel use is a straight line from minimum load to full load; there are no curved efficiency curves.

  • No parameter that keeps a unit online. A unit can be made to produce (a profile, a flow group floor), but there is no parameter that fixes its online status alone.

    A user constraint on the units online (constraint_online_coeff) sets a minimum number of units online; see How to make a unit run all the time.

Inertia and reserves

  • Inertia of units without online variables comes from their flow. A unit without online variables (no startup_method and no minimum_time_method) gives flow × inertia_constant, so a unit at half load gives half of its inertia.

    Give the synchronous units online variables (startup_method linear or binary) to count their inertia from the capacity online; see How to set a minimum inertia.

  • The inertia limit is entered, not computed. FlexTool reports the largest single flow into the group, but the inertia_limit does not follow it.

  • Upward reserve comes from online capacity. A unit with online variables holds upward reserve only from the capacity it has online. There is no separate reserve from offline units that can start fast.
  • Reserve activation is not modelled. The model holds the reserve capacity back; the energy of activating it and its cost are not part of the dispatch.

Investment and adequacy

  • The dispatch does not feed back into the investment. An investment solve does not see the results of the dispatch solves that follow it.

    If a dispatch check shows unserved energy or reserve shortfall, change the investment solve by hand: a better sample of representative periods, a capacity_margin or an energy margin, and run again; see Several milestones. The net-load re-selection of representative periods is the one loop that FlexTool runs for you.

  • The capacity margin is held by firm and VRE capacity only. Firm units count at capacity × availability, VRE at capacity × profile × availability; storage-backed units, connections and consumption count by their actual flow at the time step (see The capacity margin).

    Storage counts only with what it discharges at that step, which is conservative for a full reservoir or a long-duration store. Imports count only as their actual flow, so a group fed only by imports cannot hold a margin. A unit fed from a fuel or commodity node, or from a balance node with limited supply, counts at its capacity: the fuel is assumed to last through the peak. Connections and lossy units inside the group are left out, so the losses inside the group at the peak are not part of the requirement (add an estimate to the margin if they matter). Energy made by the balance slack of a node outside the group and exported into it is not taken back (that of a storage node feeding the group is). Under spatial Benders decomposition, a capacity-margin group with nodes in more than one region is an error.

  • Representative periods leave hours out. An investment made on a sample of weeks cannot see the hours outside the sample; check the plan with a full-year dispatch.

  • Costs count only within the model horizon. The annuity and the fixed cost of new capacity are paid in the periods it operates within the model; the years after the last period are not counted.

Discounting

  • Only inflation_rate discounts between periods. With the default 0, a cost in the last milestone weighs the same as in the first.

    In a model with real (constant-price) inputs, enter the social discount rate as inflation_rate. The per-technology discount_rate only turns investment costs into annuities; see Discounting between milestone years.

Storage

  • The standing loss is proportional to the content. self_discharge_loss is a share of the stored energy per hour. A loss that depends on the size of the storage (for example the heat loss of a hot tank) cannot be given.

User constraints and policy targets

  • Constraints on flows hold in every time step. A user constraint with flow, state or online terms is written for each time step, and its constant is a constant or a value per period, not per time step.

    There is no user constraint on a period's total flows, such as "renewables make at least 30 % of the year's electricity". The cumulative limits of a flow group (min_cumulative_flow, max_cumulative_flow) are absolute amounts (an average flow in MW), not shares: compute the amount outside FlexTool. See User constraints and Flow groups.

Networks

  • No AC power flow. A connection moves energy with losses as a share of the flow, or the network uses a linear DC power flow (dc_power_flow_with_angles). Voltages, reactive power and AC losses are not modelled.

Time, foresight and uncertainty

  • Perfect foresight within a solve. A solve knows its whole horizon, so storage and investments are planned with full knowledge of the future. Rolling and nested solves limit what each solve sees; see Solve sequences.
  • Forecasts differ only in time series. In a stochastic solve, the forecasts can differ only in the parameters marked "stochastic time" in the reference (profiles, inflow, availability and a few others); see How to use stochastics.

Mathematics and solvers

  • Linear or mixed-integer only. Every relation in the model is linear. A nonlinear effect has to be made linear (for example with several units or a profile) before it goes in.
  • Solver. The free HiGHS solver comes with FlexTool; commercial solvers can be used through Solvers. A large mixed-integer model can take long with any solver.