EOS/docs/akkudoktoreos/optimauto.md

% SPDX-License-Identifier: Apache-2.0

Automatic Optimization

Introduction

EOS offers two approaches to optimize your energy management system: post /optimize optimization and automatic optimization.

The post /optimize optimization interface, based on a POST request to /optimize, is widely used. It was originally developed by Andreas at the start of the project and is still demonstrated in his instructional videos. This interface allows users or external systems to trigger an optimization manually, supplying custom parameters and timing.

As an alternative, EOS supports automatic optimization, which runs automatically at configured intervals. It retrieves all required input data — including electricity prices, battery storage capacity, PV production forecasts, and temperature data — based on your system configuration.

Genetic Algorithm

Both optimization modes use the same core optimization engine.

EOS uses a genetic algorithm to find an optimal control strategy for home energy devices such as household loads, batteries, and electric vehicles.

In this context, each individual represents a possible solution — a specific control schedule that defines how devices should operate over time. These individuals are evaluated using resource simulations, which model the system’s energy behavior over a defined time period divided into fixed intervals.

The quality of each solution (its fitness) is determined by how well it performs during simulation, based on objectives such as minimizing electricity costs, maximizing self-consumption, or meeting battery charge targets.

Through an iterative process of selection, crossover, and mutation, the algorithm gradually evolves more effective solutions. The final result is an optimized control strategy that balances multiple system goals within the constraints of the input data and configuration.

:::{admonition} Note :class: note You don’t need to understand the internal workings of the genetic algorithm to benefit from automatic optimization. EOS handles everything behind the scenes based on your configuration. However, advanced users can fine-tune the optimization behavior using additional settings like population size, penalties, and random seed. :::

Energy Management Plan

Whenever the optimization is run, the energy management plan is updated. The energy management plan provides a list of energy management instructions in chronological order. The instructions lean on to the S2 standard to have maximum flexibility and stay completely independent from any manufacturer.

Battery Instructions

The battery control instructions assume an idealized battery model. Under this model, the battery can be operated in four discrete operation modes:

Operation Mode ID	Description
IDLE	Battery neither charges nor discharges; holds its state of charge.
CHARGE	Charge at a specified power rate up to the allowable maximum.
DISCHARGE	Discharge at a specified power rate up to the allowable maximum.
ALLOW_DISCHARGE	Allow the battery to freely discharge depending on its instantaneous power setpoint.

The operation mode factor (0.0–1.0) specifies the normalized power rate relative to the battery's nominal maximum charge or discharge power. A value of 1.0 corresponds to full-rate charging or discharging, while 0.0 indicates no power transfer. Intermediate values scale the power proportionally.

Electric Vehicle Instructions

The electric vehicle control instructions assume an idealized EV battery model. Under this model, the EV battery can be operated in two operation modes:

Operation Mode ID	Description
IDLE	Battery neither charges nor discharges; holds its state of charge.
CHARGE	Charge at a specified power rate up to the allowable maximum.

The operation mode factor (0.0–1.0) specifies the normalized power rate relative to the battery's nominal maximum charge power. A value of 1.0 corresponds to full-rate charging, while 0.0 indicates no power transfer. Intermediate values scale the power proportionally.

Home Appliance Instructions

The home appliance instructions assume an idealized home appliance model. Under this model, the home appliance can be operated in two operation modes:

Operation Mode ID	Description
RUN	The home appliance is started and runs until the end of it's power sequence.
IDLE	The home appliance does not run.

The operation mode factor (0.0–1.0) is ignored.

Configuration

Energy management configuration

The energy management is run on configured intervals with some startup delay after server start. Both values are given in seconds.

:::{admonition} Note :class: note If no interval is configured (None, null) there will be only one energy management run at startup. :::

The energy management can be run in three modes:

• OPTIMIZATION: A full optimization is done. This includes update of predictions.
• PREDICTION: Only the predictions are updated.
• DISABLED: No energy management is run.

Example:

{
    "ems": {
        "startup_delay": 5.0,
        "interval": 300.0,
        "mode": "OPTIMIZATION"
    }
}

Optimization Configuration

Optimization Time Configuration

• horizon_hours: The optimization horizon parameter defines the default time window — in hours — within which the energy optimization goal shall be achieved.

  Specific devices, like the home appliance, have their own configuration for time windows. If the time windows are not configured the simulation uses the default time window.

  Each device simulation run must ensure that all tasks or appliance cycles (e.g., running a dishwasher) are completed within the configured time windows.

• interval: Defines the time step in seconds between control actions (e.g. 3600 for one hour, 900 for 15 minutes).

:::{warning} Current Limitation

At present, the interval setting is not used by the genetic algorithm. Instead:

• The control interval is fixed to 1 hour.

Support for configurable intervals (e.g. 15-minute steps) may be added in a future release. :::

Genetic Algorithm Parameters

The behavior of the genetic algorithm can be customized using the following configuration options:

• individuals (int, default: 300): Sets the number of individuals (candidate solutions) in the (first) generation. A higher number increases solution diversity and the chance of finding a good result, but also increases computation time.

• generations (int, default: 400): Sets the number of generations to evaluate the optimal solution. In each generation, solutions are evaluated and evolved. More generations can improve optimization quality but increase computation time. Best results are usually found within a moderate number of generations.

• seed (int or null, default: null): Sets the random seed for reproducible results.

  • If null, a random seed is used (non-reproducible).

  • If an integer is provided, it ensures that the same optimization input yields the same output.

    A fixed seed to ensure reproducibility. Runs with the same seed and configuration will produce the same results.

• fitness_cache_max_entries (int or null, default: null): Limits the fitness cache of one optimization run. The cache remembers the fitness of every evaluated plan so that a repeated plan is not simulated again. It is cleared after each run, but during a run it grows with individuals × generations (roughly 100 MB at 300 × 400) while typically only 5–15 % of the evaluations hit it.

  • If null, the cache is unbounded (previous behaviour).
  • If 0, the cache is disabled; every plan is simulated.
  • If a positive integer N, the cache holds at most N keys and drops the oldest first. Each evaluated plan takes one or two keys.

  The value never changes the result: with a fixed seed, all settings give the same plan. Small devices (e.g. 512 MB RAM) should use 0 or a few thousand; this costs a few percent of run time.

• penalties (dict): Defines how penalties are applied to solutions that violate constraints (e.g., undercharged batteries). Penalty function parameter values influence the fitness score, discouraging undesirable solutions.

:::{note} Supported Penalty Functions

• ev_soc_miss: Applies a penalty when the state of charge (SOC) of the electric vehicle battery falls below the required minimum. This encourages the optimizer to ensure sufficient EV charging. Default: 10.

• ac_charge_break_even: Applies a penalty for each scheduled AC grid-charging hour where the round-trip losses (AC→DC inverter, battery internal, DC→AC inverter) mean the stored energy can never be discharged at a price that recovers the charging cost. Energy already stored in the battery from PV generation is treated as free and covers the most expensive future hours first, so the penalty only fires for the hours that remain genuinely uncovered.

  A value of 1.0 (default) means the penalty equals the actual economic loss in €. Use larger values (e.g. 3.0) to make the optimizer more aggressively avoid unprofitable AC charging, or 0.0 to disable this penalty entirely.

• battery_full_charge_miss: Applies a penalty per percent the battery stays below its target when a periodic full charge is due and enforced (see Periodic full charge). Default: 10. :::

Value Formats

• Time-related values:

  • hours: specified in hours (e.g. 24)
  • interval: specified in seconds (e.g. 3600)

• Genetic algorithm parameters:

  • individuals: must be an integer
  • seed: must be an integer or null for random behavior

• Penalty function parameter values: may be float, int, or string, depending on the type of penalty function.

Optimization configuration example

{
    "optimization": {
        "hours": 24,
        "interval": 3600,
        "genetic" : {
            "individuals": 300,
            "generations": 400,
            "seed": null,
            "penalties": {
                "ev_soc_miss": 10,
                "ac_charge_break_even": 1.0
            }
        }
    }
}

Device simulation configuration

The device simulations are used to evaluate the fitness of the individuals of the solution population.

The GENETIC algorithm supports 4 devices:

• inverter: A photovoltaic power inverter that can export to the grid and charge a battery. The inverter is mandatory.
• electric_vehicle: An electric vehicle, basically the battery of an electric vehicle. The The electrical vehicle is optional.
• battery: A battery that can be charged by the inverter. The battery is mandatory.
• home_appliance: A home appliance, like a washing machine or a dish washer. The home appliance is optional.

:::{admonition} Warning :class: warning The GENETIC algorithm can only use the first inverter, electrical vehicle, battery, home appliance that is configured, even if more devices are configured. :::

Inverter simulation configuration

Example:

{
    "devices": {
        "max_inverters": 1,
        "inverters": [
            {
                "device_id": "inv1",
                "max_power_w": 10000,
                "battery_id": "bat1",
                "ac_to_dc_efficiency": 0.95,
                "dc_to_ac_efficiency": 0.95,
                "max_ac_charge_power_w": 5000
            }
        ]
    }
}

The inverter supports separate AC↔DC conversion efficiencies:

• ac_to_dc_efficiency: Conversion loss when charging the battery from AC grid power (0-1). Set to 0 to disable AC charging. Default 1.0.
• dc_to_ac_efficiency: Conversion loss when discharging battery to AC load/grid (0-1). Must be > 0. Default 1.0.
• max_ac_charge_power_w: Maximum AC charging power in watts. null = no additional limit. Set to 0 to disable AC charging. Default null.
• dc_to_ac_efficiency_curve: Optional load-dependent DC→AC efficiency as a list of [load_fraction, efficiency] points, with the load fraction relative to max_power_w. Real inverters are much less efficient at low load, e.g. when the battery serves a small night-time base load. The points must be strictly increasing in load fraction (0-1) with efficiencies in (0, 1]; values in between are interpolated linearly, values outside are clamped to the first/last point. When set, the curve replaces dc_to_ac_efficiency for battery discharge. The efficiency is always the one at the AC power that is delivered: if the battery limits a discharge (low state of charge or its power limit), the conversion is evaluated at that lower load, not at the requested one. Default null (constant dc_to_ac_efficiency, unchanged behaviour).
• dc_to_ac_efficiency_reference_load_fraction: Load fraction at which the curve is evaluated where stored energy is valued without a specific discharge, e.g. the energy left in the battery at the end of the horizon and the AC charge break-even. Choose the typical load the battery serves. Only used with dc_to_ac_efficiency_curve. Default 0.06.

Example with an efficiency curve:

{
    "devices": {
        "inverters": {
            "inv1": {
                "max_power_w": 10000,
                "battery_id": "bat1",
                "dc_to_ac_efficiency_curve": [
                    [0.02, 0.80], [0.07, 0.93], [0.30, 0.96], [1.00, 0.95]
                ],
                "dc_to_ac_efficiency_reference_load_fraction": 0.06
            }
        }
    }
}

Electric vehicle simulation configuration

Example:

{
    "devices": {
        "max_electric_vehicles": 1,
        "electric_vehicles": {
            "ev1": {
                "device_id": "ev1",
                "battery_profile": "generic-lfp",
                "profile_adjustment": {
                    "nominal_capacity_wh": 50000,
                    "max_charge_power_w": 10000
                },
                "charge_rates": [0.0, 0.25, 0.5, 0.75, 1.0],
                "min_soc_percentage": 10,
                "max_soc_percentage": 80
            }
        }
    },
    "measurement": {
        "electric_vehicle_soc_keys": ["ev1_soc"]
    }
}

Battery simulation configuration

Example:

{
    "devices": {
        "max_batteries": 1,
        "batteries": {
            "battery1": {
                "device_id": "battery1",
                "battery_profile": "generic-lfp",
                "profile_adjustment": {
                    "nominal_capacity_wh": 8000,
                    "max_charge_power_w": 8000
                },
                "levelized_cost_of_storage_amt_kwh": 0.12,
                "min_charge_power_w": 50,
                "charge_rates": null,
                "min_soc_percentage": 5,
                "max_soc_percentage": 95
            }
        }
    }
}

Periodic full charge

LFP batteries have an almost flat voltage curve in the middle of their range. The battery management system can only recalibrate the state of charge close to full charge, so many manufacturers expect a full charge every few weeks, and some batteries start one on their own at an arbitrary time. A periodic full charge lets EOS plan it instead, preferably from PV surplus or at a cheap price.

The feature is off by default (GENETIC only). The interval and the detection threshold are properties of the battery model and come from the battery profile; the profile_adjustment of a stationary battery replaces them for this unit. The target and the use of the grid are battery settings:

• full_charge_interval_days (battery profile or profile_adjustment): maximum number of days between two full charges. null disables the feature; the bundled generic-lfp profile sets none. For LFP batteries 21 to 25 days are a reasonable start; check what the manufacturer requires.
• full_charge_detect_percentage (battery profile or profile_adjustment, profile default 99): the measured state of charge from which the battery counts as full. Many battery management systems never report exactly 100 %.
• full_charge_soc_percentage (default 100): the target. It may be above max_soc_percentage, which keeps applying on all other days, but is capped by the technical maximum SoC of the battery profile.
• full_charge_allow_grid (default false): with false, EOS only raises the maximum state of charge while the full charge is due, so that PV surplus can fill the battery; the target is not enforced. With true, the optimizer has to reach the target and charges from the grid if needed, at the cheapest time before the deadline.

The effective values are available as effective_full_charge_interval_days and effective_full_charge_detect_percentage. The cell chemistry is the chemistry of the battery profile.

The last full charge is taken from the measured state of charge of the battery (<device_id>-soc-factor), so a full charge from PV or one the battery started itself counts as well. Nothing else is stored. Without a full charge in the last interval, the interval counts from the oldest measurement, so a new installation is not forced into a full charge on its first day. The next full charge is due one interval after the last one. EOS acts once that deadline falls within the optimization horizon or has passed, and logs an INFO message for every run in which the full charge is due. The optimizer always gets at least 24 hours to finish it, so that a close or overdue full charge can still use the cheapest time.

{
    "devices": {
        "batteries": {
            "battery1": {
                "device_id": "battery1",
                "battery_profile": "generic-lfp",
                "profile_adjustment": {
                    "nominal_capacity_wh": 10000,
                    "full_charge_interval_days": 21
                },
                "max_soc_percentage": 90,
                "full_charge_allow_grid": true
            }
        }
    }
}

Home appliance simulation configuration

Example:

{
    "devices": {
        "max_home_appliances": 1,
        "home_appliances": [
            {
                "device_id": "washing machine",
                "consumption_wh": 600,
                "duration_h": 3,
                "time_windows": null,
            }
        ]
    }
}

The time windows the home appliance may run can be configured in several ways. See the time window configuration for details.

Predictions configuration

The device simulation may rely on predictions to simulate proper behaviour. E.g. the inverter needs to know the PV forecast.

Configure the predictions as described on the prediction page.

Providing your own prediction data

If EOS does not have a suitable prediction provider you can provide your own data for a prediction. Configure the respective import provider (ElecPriceImport, LoadImport, PVForecastImport, WeatherImport) and use one of the following endpoints to provide your own data:

• PUT /v1/prediction/import/ElecPriceImport
• PUT /v1/prediction/import/LoadImport
• PUT /v1/prediction/import/PVForecastImport
• PUT /v1/prediction/import/WeatherImport

Measurement configuration

Predictions and device simulations often rely on measurement data to produce accurate results. For example:

• A load forecast requires past energy meter readings.
• A battery simulation needs the current state of charge (SoC) to start from the correct condition.

Before using these features, make sure to configure the measurement as described on the measurement page.

Providing your own measurement data

You can provide your own measurement data to the prediction and simulation engine through the following REST endpoints (see the measurement page for details on the data format):

• PUT /v1/measurement/data
• PUT /v1/measurement/dataframe
• PUT /v1/measurement/series
• PUT /v1/measurement/value

Example: Supplying Battery and EV SoC

For batteries and electric vehicles, it is strongly recommended to provide current SoC. This ensures that simulations start with the correct state.

The simplest way is to use the /v1/measurement/value endpoint. Assuming the battery is named battery1 and the EV is named ev11:

1. Use the measurement keys that are pre-configured for your devices. For example:

   {
       "devices": {
           "batteries": {
               "battery1": {
               "device_id": "battery1",
               "profile_adjustment": {"nominal_capacity_wh": 8000},  ...
               "measurement_key_soc_factor": "battery1-soc-factor", ...
               }
           },
           "electric_vehicles": {
               "ev11": {
               "device_id": "ev11",
               "profile_adjustment": {"nominal_capacity_wh": 8000},  ...
               "measurement_key_soc_factor": "ev11-soc-factor", ...
               }
           }
       }
   }
   

2. Record your SoC readings to these keys.

   • Enter the values as factor of total capacity of the respective battery.

In these examples:

• datetime specifies the timestamp of the measurement.
• key is the measurement key (e.g. battery1-soc-factor).
• value is the numeric measurement value (e.g. SoC as factor of total capacity).

Raw HTTP request

PUT http://127.0.0.1:8503/v1/measurement/value?datetime=2025-09-26T16%3A39&key=battery1-soc-factor&value=0.57
PUT http://127.0.0.1:8503/v1/measurement/value?datetime=2025-09-26T16%3A39&key=ev11-soc-factor&value=0.22

Equivalent curl commands

curl -X PUT "http://127.0.0.1:8503/v1/measurement/value?datetime=2025-09-26T16%3A39&key=battery1-soc-factor&value=0.57"
curl -X PUT "http://127.0.0.1:8503/v1/measurement/value?datetime=2025-09-26T16%3A39&key=ev11-soc-factor&value=0.22"

Example: Supplying Load Data

To provide your actual load measurements in Akkudoktor-EOS:

1. Configure the measurement keys for your load energy meters. For example:

   {
       "measurements": {
           "load_emr_keys": ["my_load_meter_reading", "my_other_load_meter_reading"]
       }
   }
   

2. Record your meter readings to these keys.

   • Enter the values exactly as your energy meters report them, in kWh.
   • Use the same approach as when supplying battery or EV SoC data.