Prologue · The Assignment

Study · Model

Electricity is no longer a constant.

A cold store near Munich pays hourly spot prices. Dynamic tariffs have been mandatory since 2025 — the price beats like a restless pulse. The question: when to cool, while power is cheap and green?

As of: 2026-04-02 – 2026-07-01

01 · The Veins

Data inflow

Four streams, one organism.

Real public data flows into the system: exchange prices, generation by source, grid load, weather — hourly, across the whole window.

2,184 hours 91 days · hourly

aWATTar Exchange prices · EPEX day-ahead
Bundesnetzagentur / SMARD Generation by source, grid load
SMARD (price) Day-ahead price DE-LU
Open-Meteo / DWD Weather · temperature, radiation, wind

Data sources Public data, attributed. A model reference plant, not a client project. · CC-BY

02 · The Rhythm

Patterns

At midday the sun floods the grid.

A daily rhythm emerges: when the renewable share is high, the price falls — often below zero at midday. That is the moment to pre-cool.

Price Green power

249 hours below €0 Correlation green↔price -0.78 · 11.5 % in the window

Price €/MWh / Green power share — Sample day
h	Price	Green power %
0	127.72	30
1	119.06	33
2	115.57	30
3	113.90	28
4	109.17	27
5	105.45	29
6	98.79	38
7	80.80	55
8	53.84	78
9	18.97	99
10	3.49	100
11	0.56	100
12	0.82	100
13	0.00	100
14	0.98	100
15	11.31	95
16	52.44	78
17	100.32	57
18	129.39	35
19	143.07	19
20	155.80	10
21	156.71	8
22	151.37	10
23	142.58	13

03 · Looking Ahead

Forecast

The model extends the curve.

A gradient-boosting model forecasts price and an uncertainty band for tomorrow — measured honestly in a walk-forward backtest, including where it misses.

Actual Forecast P10–P90

16.77 €/MWh MAE Naive 51.91 · Persistence 31.03

Error per hour ahead (h1–24)

Actual / Forecast €/MWh — Sample day
h	Actual	Forecast	P10	P90
0	127.72	122.47	114.34	129.12
1	119.06	113.08	109.57	127.33
2	115.57	112.58	109.22	126.70
3	113.90	117.72	111.55	123.98
4	109.17	117.65	111.78	125.90
5	105.45	117.37	111.05	120.35
6	98.79	103.62	102.32	109.00
7	80.80	90.40	84.59	101.33
8	53.84	49.42	17.95	70.96
9	18.97	1.68	-3.62	24.28
10	3.49	-3.85	-38.03	11.52
11	0.56	-20.12	-43.17	6.23
12	0.82	-16.37	-43.40	6.18
13	0.00	-16.69	-50.02	3.28
14	0.98	-1.37	-6.00	1.81
15	11.31	-1.98	-8.46	14.11
16	52.44	36.41	21.88	51.46
17	100.32	67.51	65.20	96.73
18	129.39	112.69	109.97	122.67
19	143.07	130.20	126.03	139.21
20	155.80	150.35	133.81	159.96
21	156.71	149.64	136.58	151.84
22	151.37	137.45	132.90	143.05
23	142.58	128.83	124.15	134.32

04 · The Decision

Load plan

Cool when it’s cheap.

The refrigeration plant becomes a thermal battery: pre-cool in the cheap, green hours, coast through the expensive ones. An optimiser sets the schedule — against the naive baseline.

cheap hours Temperature

€2.21 saved · day naive €10.78 · optimal €8.57

Cooling load kW / Temperature °C — Sample day
h	Cooling load	Temperature
0	11.9	-18.0
1	11.7	-18.0
2	11.7	-18.0
3	11.6	-18.0
4	11.3	-18.0
5	11.2	-18.0
6	11.1	-18.0
7	11.7	-18.0
8	12.7	-18.0
9	13.8	-18.0
10	14.6	-18.0
11	36.9	-19.1
12	0.0	-18.3
13	50.0	-20.0
14	16.8	-20.0
15	17.1	-20.0
16	17.4	-20.0
17	17.5	-20.0
18	17.2	-20.0
19	8.2	-19.6
20	0.0	-18.8
21	0.0	-18.0
22	14.2	-18.0
23	13.6	-18.0

05 · The Voice

AI briefing

The system speaks.

From the load plan’s numbers, a local language model shapes a German briefing — grounded in real values, no hallucination.

Laden Sie die Kälteanlage in den günstigen Stunden 11 bis 14 auf, um 2 EUR (20 %) zu sparen. Nutzen Sie dabei den hohen VRE-Anteil von 53 % für eine effiziente Betriebsführung. System schlägt vor, der Mensch entscheidet.

The system’s own words — in German, grounded in the day’s numbers.

Local Qwen3.6 · the data never leaves the machine

06 · The Reflex

Automation

Every day, by itself.

Around 14:00 tomorrow’s prices arrive — the system computes the plan and reports in. A reflex, not an autopilot.

1 ≈ 14:00
2 Tomorrow’s prices
3 Compute plan
4 Alert
↻ daily

The system proposes — the human decides.

★ · The Proof

The result

Summed across the whole window.

No single-day luck: the savings add up over dozens of days. That is the proof — one operator, full stack, on a real system.

Over 89 days €249.75 saved · −18.4 % naive €1,359.08 · optimal €1,109.32

cumulative savings

Daily savings € — 89 days
Date	€
2026-04-03	1.44
2026-04-04	2.24
2026-04-05	2.22
2026-04-06	3.80
2026-04-07	3.43
2026-04-08	3.41
2026-04-09	2.35
2026-04-10	2.25
2026-04-11	2.43
2026-04-12	2.27
2026-04-13	1.20
2026-04-14	1.80
2026-04-15	2.08
2026-04-16	2.20
2026-04-17	2.10
2026-04-18	2.09
2026-04-19	1.82
2026-04-20	1.64
2026-04-21	2.30
2026-04-22	2.68
2026-04-23	2.88
2026-04-24	2.76
2026-04-25	4.20
2026-04-26	9.25
2026-04-27	2.99
2026-04-28	2.37
2026-04-29	3.31
2026-04-30	3.10
2026-05-01	10.62
2026-05-02	4.65
2026-05-03	2.26
2026-05-04	3.04
2026-05-05	1.10
2026-05-06	0.97
2026-05-07	1.67
2026-05-08	1.33
2026-05-09	2.09
2026-05-10	1.89
2026-05-11	0.72
2026-05-12	1.90
2026-05-13	1.40
2026-05-14	2.05
2026-05-15	1.51
2026-05-16	2.04
2026-05-17	2.21
2026-05-18	1.89
2026-05-19	1.39
2026-05-20	2.01
2026-05-21	2.93
2026-05-22	3.02
2026-05-23	3.00
2026-05-24	3.42
2026-05-25	3.00
2026-05-26	3.29
2026-05-27	2.44
2026-05-28	5.79
2026-05-29	3.28
2026-05-30	2.46
2026-05-31	2.20
2026-06-01	3.97
2026-06-02	1.04
2026-06-03	1.69
2026-06-04	1.79
2026-06-05	2.50
2026-06-06	2.25
2026-06-07	2.74
2026-06-08	2.17
2026-06-09	2.41
2026-06-10	1.85
2026-06-11	1.90
2026-06-12	1.06
2026-06-13	2.18
2026-06-14	1.99
2026-06-15	2.49
2026-06-16	1.82
2026-06-17	1.33
2026-06-18	6.27
2026-06-19	2.72
2026-06-20	2.60
2026-06-21	2.18
2026-06-22	3.51
2026-06-23	7.19
2026-06-24	9.00
2026-06-25	3.66
2026-06-26	4.40
2026-06-27	3.19
2026-06-28	2.28
2026-06-29	3.36
2026-06-30	6.02

Your turn.

Pick a day, scrub the price curve, toggle the optimal plan on and off — the cost recomputes in front of you.

Frozen window 2026-04-02–2026-07-01 no live data

Day in the window

Price curve Forecast from 2026-05-09

Price Green power Forecast below €0

12:00 Price 0.82 €/MWh Green power 100 % Forecast -9 ±25 €/MWh

Cooling plan

Naive €10.78 Plan €8.57 Savings −€2.21

Questions for the operation Pre-generated by the local Qwen3.6 — no live requests.

Wann liegt typischerweise das günstigste Zeitfenster im Tag?

Das günstigste Zeitfenster liegt typischerweise zwischen 12 und 15 Uhr.

Warum sind die Strompreise mittags oft negativ?

Mittags flutet Solarstrom das Netz, wodurch die Preise fallen und teils negativ werden.

Wie viel spart die optimierte Vorkühlung im gesamten Zeitfenster?

Die optimierte Vorkühlung spart im gesamten Zeitfenster 249,75 EUR.

Wie zuverlässig ist die Preisprognose im Vergleich zur Naiv-Methode?

Die Prognose ist deutlich zuverlässiger als die Naiv-Methode. Der mittlere absolute Fehler (MAE) beträgt 16,77 im Vergleich zu 51,91 der Naiv-Methode, was einer Reduktion um rund 68 % entspricht.

Replay the daily briefing

Laden Sie die Kälteanlage in den günstigen Stunden 11 bis 14 auf, um 2 EUR (20 %) zu sparen. Nutzen Sie dabei den hohen VRE-Anteil von 53 % für eine effiziente Betriebsführung. System schlägt vor, der Mensch entscheidet.

Price €/MWh / Green power % — 2026-04-02–2026-07-01
h	Price	Green power %
0	127.72	30
1	119.06	33
2	115.57	30
3	113.90	28
4	109.17	27
5	105.45	29
6	98.79	38
7	80.80	55
8	53.84	78
9	18.97	99
10	3.49	100
11	0.56	100
12	0.82	100
13	0.00	100
14	0.98	100
15	11.31	95
16	52.44	78
17	100.32	57
18	129.39	35
19	143.07	19
20	155.80	10
21	156.71	8
22	151.37	10
23	142.58	13

Forecast €/MWh — P10 / P50 / P90
h	P10	P50	P90
0	114.34	118.96	129.12
1	109.57	115.57	127.33
2	109.22	115.11	126.70
3	111.55	114.73	123.98
4	111.78	113.53	125.90
5	111.05	111.42	120.35
6	102.32	103.85	109.00
7	84.59	92.94	101.33
8	17.95	50.06	70.96
9	-3.62	-3.47	24.28
10	-38.03	-8.74	11.52
11	-43.17	-9.65	6.23
12	-43.40	-9.23	6.18
13	-50.02	-11.19	3.28
14	-6.00	0.97	1.81
15	-8.46	3.71	14.11
16	21.88	50.32	51.46
17	65.20	95.96	96.73
18	109.97	112.47	122.67
19	126.03	138.01	139.21
20	133.81	151.41	159.96
21	136.58	150.73	151.84
22	132.90	141.02	143.05
23	124.15	134.12	134.32

Savings € — Day 1–89
Day	Price €
1	1.44
2	2.24
3	2.22
4	3.80
5	3.43
6	3.41
7	2.35
8	2.25
9	2.43
10	2.27
11	1.20
12	1.80
13	2.08
14	2.20
15	2.10
16	2.09
17	1.82
18	1.64
19	2.30
20	2.68
21	2.88
22	2.76
23	4.20
24	9.25
25	2.99
26	2.37
27	3.31
28	3.10
29	10.62
30	4.65
31	2.26
32	3.04
33	1.10
34	0.97
35	1.67
36	1.33
37	2.09
38	1.89
39	0.72
40	1.90
41	1.40
42	2.05
43	1.51
44	2.04
45	2.21
46	1.89
47	1.39
48	2.01
49	2.93
50	3.02
51	3.00
52	3.42
53	3.00
54	3.29
55	2.44
56	5.79
57	3.28
58	2.46
59	2.20
60	3.97
61	1.04
62	1.69
63	1.79
64	2.50
65	2.25
66	2.74
67	2.17
68	2.41
69	1.85
70	1.90
71	1.06
72	2.18
73	1.99
74	2.49
75	1.82
76	1.33
77	6.27
78	2.72
79	2.60
80	2.18
81	3.51
82	7.19
83	9.00
84	3.66
85	4.40
86	3.19
87	2.28
88	3.36
89	6.02