Branch 5: dynamický terminal SoC factor při future neg buy

backend/services/planning_engine.py

@@ -91,8 +91,11 @@ NEG_EVENING_PREP_DISCHARGE_SHORTFALL_PENALTY_CZK_KWH = 120.0
 # Kotva: SoC na konci večera D−1 a těsně před 1. sell<0 ráno D ≤ reserve_soc.
 NEG_EVENING_RESERVE_SOC_MAX_SLACK_WH = 400.0
 NEG_EVENING_RESERVE_SOC_SLACK_PENALTY_CZK_PER_WH = 55.0
-# Terminal SoC shadow price: při blízkém buy<0 nesmí LP „šetřit“ baterii ve večerní špičce.
-FUTURE_NEG_BUY_TERMINAL_SOC_FACTOR_MULT = 0.1
+# Terminal SoC shadow price: effective_factor = base × (1 − w_neg); w_neg roste s blízkostí a záporností buy<0.
+TERMINAL_NEG_BUY_WEIGHT_HORIZON_SLOTS = int(36 / INTERVAL_H)
+TERMINAL_NEG_BUY_MAGNITUDE_REF_CZK = 1.0
+TERMINAL_NEG_BUY_MAGNITUDE_FLOOR = 0.25
+TERMINAL_NEG_BUY_WEIGHT_CAP = 0.95
 # Před prvním sell<0: export FVE jen pokud predikce v sell<0 okně pokryje dobítí na prep cíl.
 PRE_NEG_PV_EXPORT_FORECAST_MARGIN = 1.15
 PRE_NEG_PV_EXPORT_MIN_NEEDED_WH = 2500.0
@@ -1388,6 +1391,38 @@ def _first_neg_sell_idx_on_prague_day(
     return None
 
 
+def _terminal_neg_buy_weight(
+    slots: list[PlanningSlot],
+    *,
+    first_neg_buy_idx: int | None,
+) -> float:
+    """
+    w_neg ∈ [0, TERMINAL_NEG_BUY_WEIGHT_CAP]: snížení terminal SoC shadow price při buy<0 v horizontu.
+    Blížší a zápornější okno → vyšší váha; effective_factor = planner_terminal_soc_value_factor × (1 − w_neg).
+    """
+    if first_neg_buy_idx is None or first_neg_buy_idx <= 0:
+        return 0.0
+    slots_ahead = first_neg_buy_idx
+    prox = max(
+        0.0,
+        1.0 - slots_ahead / TERMINAL_NEG_BUY_WEIGHT_HORIZON_SLOTS,
+    )
+    if prox <= 0.0:
+        return 0.0
+    window_end = min(len(slots), first_neg_buy_idx + int(24 / INTERVAL_H))
+    neg_buys = [
+        float(slots[t].buy_price)
+        for t in range(first_neg_buy_idx, window_end)
+        if float(slots[t].buy_price) < 0.0
+    ]
+    if not neg_buys:
+        return 0.0
+    min_neg_buy = min(neg_buys)
+    mag_raw = min(1.0, abs(min_neg_buy) / TERMINAL_NEG_BUY_MAGNITUDE_REF_CZK)
+    mag = TERMINAL_NEG_BUY_MAGNITUDE_FLOOR + (1.0 - TERMINAL_NEG_BUY_MAGNITUDE_FLOOR) * mag_raw
+    return min(TERMINAL_NEG_BUY_WEIGHT_CAP, prox * mag)
+
+
 def _future_neg_buy_discharge_enabled(
     slots: list[PlanningSlot],
     battery: Any,
@@ -2802,9 +2837,11 @@ def solve_dispatch(
                 neg_sell_soc_target_by_t if neg_sell_phases_en else None
             ),
         )
-    terminal_factor = terminal_factor_base
-    if future_neg_buy_discharge_en:
-        terminal_factor *= FUTURE_NEG_BUY_TERMINAL_SOC_FACTOR_MULT
+    terminal_neg_buy_weight = _terminal_neg_buy_weight(
+        slots,
+        first_neg_buy_idx=first_neg_buy_idx,
+    )
+    terminal_factor = terminal_factor_base * (1.0 - terminal_neg_buy_weight)
     # Kč/Wh: ocenění energie ponechané v baterii na konci horizontu (receding horizon kotva).
     terminal_soc_kcz_per_wh = avg_buy_terminal * terminal_factor / 1000.0
 
@@ -4903,6 +4940,7 @@ def solve_dispatch(
             ),
             "evening_push_hard_suppressed": bool(evening_push_hard_suppressed),
             "future_neg_buy_discharge": bool(future_neg_buy_discharge_en),
+            "terminal_neg_buy_weight": float(terminal_neg_buy_weight),
             "terminal_soc_factor_effective": float(terminal_factor),
             "pos_sell_pre_neg_buy_ge_exempt_slots": [
                 slots[i].interval_start.isoformat()

backend/tests/test_planning_dispatch_milp.py

@@ -42,6 +42,7 @@ from services.planning_engine import (
     _slots_until_buy_le_threshold,
     _slots_until_sell_lt,
     _soc_panel_min_wh_series,
+    _terminal_neg_buy_weight,
     solve_dispatch,
     solve_dispatch_two_pass,
 )
@@ -2717,6 +2718,113 @@ class TerminalSocShadowTests(unittest.TestCase):
             msg="terminal SoC shadow price should keep end-of-horizon SoC above bare minimum",
         )
 
+    def test_terminal_neg_buy_weight_zero_without_future_neg_buy(self) -> None:
+        base = datetime(2026, 4, 3, 12, 0, tzinfo=timezone.utc)
+        slots = [
+            PlanningSlot(
+                interval_start=base + timedelta(minutes=15 * i),
+                buy_price=2.0,
+                sell_price=3.0,
+                pv_a_forecast_w=0,
+                pv_b_forecast_w=0,
+                load_baseline_w=600,
+                ev1_connected=False,
+                ev2_connected=False,
+            )
+            for i in range(8)
+        ]
+        self.assertEqual(_terminal_neg_buy_weight(slots, first_neg_buy_idx=None), 0.0)
+        self.assertEqual(_terminal_neg_buy_weight(slots, first_neg_buy_idx=0), 0.0)
+
+    def test_terminal_neg_buy_weight_scales_with_proximity_and_magnitude(self) -> None:
+        base = datetime(2026, 6, 6, 19, 0, tzinfo=timezone.utc)
+        slots_far: list[PlanningSlot] = []
+        for i in range(96):
+            buy = -0.9 if i >= 64 else 3.0
+            slots_far.append(
+                PlanningSlot(
+                    interval_start=base + timedelta(minutes=15 * i),
+                    buy_price=buy,
+                    sell_price=5.0 if buy > 0 else -0.2,
+                    pv_a_forecast_w=0,
+                    pv_b_forecast_w=0,
+                    load_baseline_w=500,
+                    ev1_connected=False,
+                    ev2_connected=False,
+                )
+            )
+        slots_near: list[PlanningSlot] = []
+        for i in range(32):
+            buy = -1.0 if i >= 8 else 3.0
+            slots_near.append(
+                PlanningSlot(
+                    interval_start=base + timedelta(minutes=15 * i),
+                    buy_price=buy,
+                    sell_price=5.0 if buy > 0 else -0.2,
+                    pv_a_forecast_w=0,
+                    pv_b_forecast_w=0,
+                    load_baseline_w=500,
+                    ev1_connected=False,
+                    ev2_connected=False,
+                )
+            )
+        w_far = _terminal_neg_buy_weight(slots_far, first_neg_buy_idx=64)
+        w_near = _terminal_neg_buy_weight(slots_near, first_neg_buy_idx=8)
+        self.assertGreater(w_far, 0.2)
+        self.assertLess(w_far, 0.75)
+        self.assertGreater(w_near, w_far)
+        self.assertGreater(w_near, 0.8)
+
+    def test_terminal_soc_factor_reduced_when_future_neg_buy(self) -> None:
+        prague = ZoneInfo("Europe/Prague")
+        base = datetime(2026, 6, 6, 19, 0, tzinfo=prague).astimezone(timezone.utc)
+        slots: list[PlanningSlot] = []
+        for i in range(96):
+            local = (base + timedelta(minutes=15 * i)).astimezone(prague)
+            d, h = local.day, local.hour
+            if d == 6:
+                buy, sell = 3.0, 5.3
+            else:
+                buy = -0.89 if 11 <= h < 14 else 2.0
+                sell = -0.2 if 5 <= h < 15 else 2.5
+            slots.append(
+                PlanningSlot(
+                    interval_start=base + timedelta(minutes=15 * i),
+                    buy_price=buy,
+                    sell_price=sell,
+                    pv_a_forecast_w=0,
+                    pv_b_forecast_w=0,
+                    load_baseline_w=500,
+                    ev1_connected=False,
+                    ev2_connected=False,
+                    allow_charge=True,
+                    allow_discharge_export=True,
+                )
+            )
+        battery = _battery(uc_wh=64_000.0, terminal_soc_value_factor=0.9)
+        hp = SimpleNamespace(rated_heating_power_w=0, tuv_min_temp_c=45.0, tuv_target_temp_c=55.0)
+        grid = SimpleNamespace(max_import_power_w=17_000, max_export_power_w=13_500)
+        vehicles = [
+            SimpleNamespace(max_charge_power_w=0, battery_capacity_kwh=1.0, default_target_soc_pct=80.0),
+            SimpleNamespace(max_charge_power_w=0, battery_capacity_kwh=1.0, default_target_soc_pct=80.0),
+        ]
+        _results, _ms, snap = solve_dispatch(
+            slots,
+            battery,
+            hp,
+            grid,
+            [None, None],
+            vehicles,
+            0.5 * battery.soc_max_wh,
+            50.0,
+            operating_mode="AUTO",
+        )
+        w_neg = float(snap["inputs"]["terminal_neg_buy_weight"])
+        eff = float(snap["inputs"]["terminal_soc_factor_effective"])
+        self.assertGreater(w_neg, 0.3)
+        self.assertLess(eff, 0.6)
+        self.assertAlmostEqual(eff, 0.9 * (1.0 - w_neg), places=6)
+
 
 class SpreadGuardHome01EconomicsTests(unittest.TestCase):
     """Regrese: sell≪buy (VT) nesmí vést k PV exportu + masivnímu grid importu ve stejném slotu."""