Optimization Strategy¶

This page explains the high-level optimization strategy and how to tune the optimizer for your specific situation.

Optimization Objectives¶

The integration optimizes a multi-objective problem with the following goals:

Primary Objective: Minimize Cost¶

Minimize total electricity cost while meeting heat demand.

Constraints¶

1. Heat balance: All heat demand must be satisfied (or buffered/deferred)
2. Temperature bounds: Supply temperature within limits
3. Rate limits: Maximum ±1°C offset change per hour
4. Heat debt limit: Buffer ≥ -max_buffer_debt (allows negative buffer)

Implicit Objectives¶

• Comfort: Maintained by temperature bounds
• System longevity: Protected by rate limits
• Efficiency: Naturally optimized by COP model

Optimization Strategies¶

1. Temporal Shifting (Heat Debt Enabled)¶

Concept: Heat when electricity is cheap, allow temperature reduction during expensive periods.

Mechanism:

• Pre-heat during low-price periods (+offset, builds positive buffer)
• Under-heat during high-price periods (-offset, creates heat debt)
• Heat debt must be repaid within planning horizon

Two buffer states:

• Positive buffer: Excess heat from pre-heating or solar gain
• Negative buffer (heat debt): Allowed temperature reduction during expensive hours

Effectiveness:

• High with dynamic pricing (15-30% savings possible)
• Low with fixed pricing (only COP optimization)
• Requires configurable max_buffer_debt (default: 5.0 kWh)

2. COP Optimization¶

Concept: Use lower supply temperatures (higher COP) when possible.

Mechanism:

• Lower offset → Lower supply temp → Higher COP → Less electricity
• Trade-off: Less heat output vs. better efficiency

Effectiveness:

• Moderate (5-10% improvement)
• Works even with fixed pricing

Example:

Lower temp uses less electricity despite COP gain.

3. Solar Integration¶

Concept: Time heating to coincide with solar production and gain.

Mechanism:

• Reduce heating during solar gain (natural warming)
• If production > consumption, heat during excess (self-consumption)
• Buffer accumulates from both solar gain and pre-heating

Effectiveness:

• Very high with solar (30-70% on optimal days)
• Seasonal (best in spring/fall)

4. Predictive Buffering¶

Concept: Build buffer in anticipation of high prices or low temperatures.

Mechanism:

• Forecast identifies expensive periods
• Pre-heat before expensive periods
• Buffer carries through expensive period

Effectiveness:

• Depends on forecast accuracy (good with day-ahead pricing)
• Limited by buffer capacity (thermal mass)

Trade-offs and Balancing¶

COP vs. Timing¶

The optimizer must balance:

• High COP (low supply temp, low offset) → efficient but may occur during expensive period
• Low COP (high supply temp, high offset) → inefficient but during cheap period

Decision factor:

If price ratio > COP ratio → favor timing over efficiency

Example:

• Cheap period: €0.15/kWh at COP 3.5 (40°C)
• Expensive period: €0.40/kWh at COP 4.0 (35°C)

Cheap cost: €0.15 / 3.5 = €0.043/kWh_heat

Expensive cost: €0.40 / 4.0 = €0.100/kWh_heat

Verdict: Heat during cheap period (2.3× lower cost per heat)

Comfort vs. Cost¶

While minimizing cost, the optimizer respects comfort:

• Temperature bounds: Prevent too-cold or too-hot supply temps
• Rate limits: Avoid rapid temperature swings
• Heat debt limit: Buffer ≥ -max_buffer_debt prevents excessive cooling

The optimizer can create "heat debt" (negative buffer) up to the configured limit. This means:

• Indoor temperature may drop slightly during expensive hours
• Temperature is restored during cheaper hours
• Comfort impact: Depends on max_buffer_debt setting
• 0 kWh = No debt, always meet demand immediately (conservative)
• 5.0 kWh = Default, moderate debt (~1-2°C variation possible)
• 10+ kWh = Aggressive, larger temperature swings

Recommendation: Start with default (5.0 kWh) and monitor comfort. Reduce if too cold during expensive hours, increase if willing to accept more variation for higher savings.

Forecast Horizon vs. Computation¶

Longer planning windows allow better optimization but increase computation:

Recommendation: 6 hours is optimal for most users

Tuning Parameters¶

1. Planning Window¶

Location: Configuration → Advanced Settings

Default: 6 hours

Recommendations:

• 3-4 hours: Low-power devices, simpler scenarios
• 6 hours: Default, good balance
• 12 hours: High price volatility, predictable patterns
• 24 hours: Maximum optimization, powerful hardware

2. Time Base¶

Location: Configuration → Advanced Settings

Default: 60 minutes

Recommendations:

• 60 minutes: Default, matches hourly pricing
• 30 minutes: Half-hourly pricing markets
• 15 minutes: Expert users, high-resolution control

3. Temperature Bounds¶

Location: Number entities

Defaults:

• Min supply temp: 25°C
• Max supply temp: 50°C

Tuning:

• Tighter bounds (e.g., 30-45°C): Less optimization flexibility, more comfort
• Wider bounds (e.g., 25-55°C): More optimization flexibility, verify heat pump supports range

4. K-Factor Calibration¶

Impact: Directly affects COP predictions and optimization aggressiveness

Calibration process:

1. Monitor actual heat pump COP at different supply temps
2. Plot COP vs. supply temp
3. Calculate slope: k ≈ ΔCOP / ΔT_supply
4. Adjust in configuration

Example:

K-factor too low → increase to 0.04

Re-test:

Much better!

5. COP Compensation Factor¶

Impact: Overall scaling of efficiency predictions

Calibration:

1. Measure total electricity consumption over 24 hours
2. Measure total heat delivered (or calculate from degree-days)
3. Calculate actual system COP = heat / electricity
4. Compare to theoretical COP from model
5. Adjust compensation factor

Example:

• Theoretical average COP: 3.8
• Measured system COP: 3.4
• Compensation factor: 3.4 / 3.8 = 0.89

Advanced Optimization Techniques¶

1. Price Forecast Extrapolation¶

When next-day prices are unavailable, the integration:

1. Uses current price as constant forecast
2. Applies historical volatility patterns (if configured)
3. Falls back to COP-only optimization

2. Weather Forecast Refinement¶

The integration fetches weather from open-meteo.com:

• Updates every 5 minutes
• Uses 24-hour forecast
• Automatically adjusts for forecast errors (via re-optimization)

3. Adaptive Buffer Management¶

Buffer capacity is not explicitly limited, but the optimizer:

• Tracks buffer evolution in state space
• Prevents negative buffer (cannot borrow heat)
• Naturally saturates when thermal mass is full

4. Constraint Relaxation (Future Feature)¶

Currently under consideration:

• Soft temperature bounds: Violate bounds with penalty
• Variable rate limits: Allow faster changes during emergencies
• Demand response: Reduce heating during grid stress

Optimization Performance Metrics¶

The integration exposes several metrics to monitor optimization quality:

1. Diagnostics Sensor¶

sensor.heating_curve_optimizer_diagnostics

Attributes include:

• optimization_time_ms: How long optimization took
• states_explored: Number of DP states
• optimal_cost: Predicted cost for planning window
• buffer_peak: Maximum buffer in forecast

2. COP Efficiency Delta¶

sensor.heating_curve_optimizer_cop_efficiency_delta

Shows COP improvement from optimization:

Positive values indicate efficiency gains.

3. Heat Generation Delta¶

sensor.heating_curve_optimizer_heat_generation_delta

Shows heat output difference:

Real-World Optimization Patterns¶

Pattern 1: Morning Pre-Heat¶

Offset
  +4 │ ╱╲
     │╱  ╲___
   0 │        ╲___
     │            ╲___
  -4 │                ╲___
     └───────────────────────
      06  09  12  15  18  Time

Price
High │        ████████
     │    ████        ████
Low  │████                ████
     └───────────────────────
      06  09  12  15  18  Time

Strategy: Pre-heat during low morning prices, coast during high midday prices

Pattern 2: Solar Sync¶

Offset
  +4 │
     │
   0 │╲___
     │    ╲___________╱╲
  -4 │        ▼
     └───────────────────────
      06  09  12  15  18  Time

Solar
High │        ╱╲
     │      ╱    ╲
Low  │___╱          ╲___
     └───────────────────────
      06  09  12  15  18  Time

Strategy: Reduce heating during solar peak, resume as solar fades

Pattern 3: Cold Snap Response¶

Offset
  +4 │██████████████████
     │
   0 │
     │
  -4 │
     └───────────────────────
      00  06  12  18  24  Time

Temp
 10°│
   0│
 -10│___________________
     └───────────────────────
      00  06  12  18  24  Time

Strategy: Maximum offset throughout extreme cold (optimization limited by capacity)

Troubleshooting Poor Optimization¶

Symptom: Offset Always Zero¶

Causes:

• No price forecast available
• All prices equal (fixed pricing)
• Very cold weather (at capacity limits)

Solutions:

• Verify price sensor has forecast attributes
• Consider dynamic pricing contract
• Accept limited optimization in extreme weather

Symptom: Excessive Offset Changes¶

Causes:

• Incorrect k-factor (too low)
• No rate limiting (shouldn't happen)
• Buggy price forecast

Solutions:

• Increase k-factor
• Check price sensor data quality
• Enable rate limiting (should be default)

Symptom: High Buffer, No Utilization¶

Causes:

• Offset constraints too tight
• Min supply temp too high
• Optimization not accounting for buffer

Solutions:

• Lower min supply temp (e.g., 25°C instead of 35°C)
• Widen offset range
• Check diagnostics for buffer tracking

Related:

• Dynamic Programming - Algorithm internals
• COP Calculation - Efficiency modeling
• Buffer System - Thermal storage