Automation Decision Network

[ericdmoore]

Expected Schedule

Thermostat

Heating

• Solar Heater Panel
• Water Heater Elements

Cooling

• Night Time Pre Cooling
• Day Time Air Circulation
• Inlet + Exhaust Fans

Water Cleaning

• Water Pump

Lights

Interior + Exterior

Constraint Based Planning

• Todays Work -vs- Todays Energy Forecast

• Tomorrows Work -vs- Energy Forecast Some work is a function of ambient air temp. So might wish to pull that in too.

• Tomo^2 Work -vs- Energy Forecast

• ...possibly looks out further

Base Cases

• Today: Clear Sky, next 5 days - cloudy

• Today: Cloudy; next 5: clear

• Today: Clear; next 5: alternating cloudy/clear

[ericdmoore]

Perhaps want a heat, cool, collar mode?

Prioritize:

• Thermostat Activities
• Water
• Lighting Circuits
• State of Charge Margin

The decisions seem to really center on:

• "When to turn on" and
• "how long to run" (when to turn off)

[ericdmoore]

Goal Stack

• Don't go offline (meaning low voltage cutoff)
• Achieve as many scheduled tasks as permitted based on state, forecast, and plan.
• I suppose you could automate a step higher and say keep cold plunge clean.

Rough Estimate of the meta algorithm

• each time step evaluate the following
• possible actions
• possible inactions
• possible messages

Some goals might be to

• maximize the time and degree above/below ambient temp.
  • potentially only during sun hours
• or to achieve a min run time of X and bonus run time thereafter

[ericdmoore]

Seems related to:

• https://en.wikipedia.org/wiki/Optimal_control

And potentially other types of control theory.

Consider looking for python packages or js packages that implement some type of solver for this style of problem

[ericdmoore]

See #2 as related inputs and outputs

[ericdmoore]

Heating alone is subject to PID controller style control problems.

[ericdmoore]

Constraints

• a model of energy consumption per item
• a model of potential energy production and timings
• high frequency cut off for switching. Aka once you turn on a switch - leave it on for N seconds
• a response model for objective(s) what if this is what we don't know or don't have. Can we start with something terrible? And update the model? How often should the model update?
• a delay model per item - for example water will take 1s to flow up the pipe and into the heat exchangers

[ericdmoore]

Isn't this a feedback loop with inputs of:

State(T-1) + ActionSet = State(T)

Then repeat indefinitely?

Kinda - it does not incorporate loss functions or maximization efforts to move the state towards an objective.

[ericdmoore]

// Assume functions for sensing, prediction, and actuation are defined elsewhere function senseCurrentState() { // Returns current system state, e.g., temperatures, power levels }

function predictFutureStates(currentState, forecasts) { // Use models to predict future states based on current state and forecasts (weather, solar) }

function determineOptimalActions(predictedStates) { // Determine the best actions to take based on predicted states to achieve objectives // This could involve solving an optimization problem }

function actuateControls(actions) { // Apply the determined actions to the system, e.g., turning relays on/off }

function mainControlLoop() { while (true) { // Main loop const currentState = senseCurrentState(); // Get current system state const forecasts = getForecasts(); // Assume this function gets weather and solar forecasts const predictedStates = predictFutureStates(currentState, forecasts); // Predict future states const optimalActions = determineOptimalActions(predictedStates); // Determine optimal actions actuateControls(optimalActions); // Actuate controls based on optimal actions

// Wait for a specified interval before next loop iteration
// This wait time depends on the specifics of your system and control requirements
sleep(timeInterval);

} }

// Start the main control loop mainControlLoop();

[ericdmoore]

Determine optimal actions is the same as saying minimize error or maximize objective.

[ericdmoore]

Perhaps the objective function includes terms like:

maximize the inside air degree-minutes below outside-ambient temperature.

But in actuality if it gave up on cooling during the hottest part of the day, and sand-bagged during the cooler parts - this would be an unsatisfactory outcome.

So potentially it's more like minimize the accumulated degree-minutes. Or Maximize the difference in projected vs actual accumulated degree minutes.

[ericdmoore]

Predict future states - requires some model of what various actions will do to the state.

And model updating can happen after the sleep timer.

Where updatemodel(currentState, actionsToStart, actuallyHappenedState) can all happen after the sleep and by sensing the state before the loop starts again.

Perhaps the sensor can be proxied through a memoize function to save time and battery.