Load ML Improvements, Put back GivTCP Load by springfall2008 · Pull Request #3392 · springfall2008/batpred

springfall2008 · 2026-02-19T08:50:40Z

No description provided.

Copilot

Pull request overview

Updates Predbat’s example dashboard charts to include ML-based home load/power forecasting visuals, bumps the app version, and adjusts how inverter load power is derived from GivTCP REST data.

Changes:

Add two new ApexCharts examples for ML Home Load Forecast and ML Home Power Forecast (excluding EV) to the example chart template.
Bump Predbat version to v8.33.4.
Switch GivTCP REST load power source to Load_Power (instead of energy-balance-derived load).

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 6 comments.

File	Description
templates/example_chart.yml	Adds ML load/power forecast chart examples and related JS data_generators.
apps/predbat/predbat.py	Version bump to `v8.33.4`.
apps/predbat/inverter.py	Changes REST-derived `load_power` assignment to use `Load_Power`.

Comments suppressed due to low confidence (1)

apps/predbat/inverter.py:1108

Leftover commented-out debug logging should be removed or gated behind a real debug flag/config (commented code tends to accumulate and makes this hot path harder to read).

                # self.log("DEBUG: Power details from REST: {}".format(ppdetails))

Copilot · 2026-02-19T08:54:48Z

templates/example_chart.yml

+    attribute: power_today_now
+    name: Home Power (Avg 1h)
+    type: line
+    yaxis_id: forecasts
+    color: black
+    stroke_width: 3
+    curve: smooth
+    opacity: 1
+    extend_to: now
+    show:
+      in_header: false
+      in_legend: false
+    group_by:
+      func: avg
+      duration: 6h


The label says "Home Power (Avg 1h)", but group_by.duration is set to 6h, so the displayed series won’t match the name. Either change the name or adjust the group_by duration to align with the label.

Copilot · 2026-02-19T08:54:49Z

apps/predbat/inverter.py

-                # Calculate load from energy balance instead of using inverter register (which is incorrect during grid charging)
-                # Load = PV + Grid + Battery (battery negative when charging, positive when discharging)
-                self.load_power = self.pv_power + self.grid_power + self.battery_power
+                self.load_power = float(ppdetails.get("Load_Power", 0.0))


Reading REST "Load_Power" directly changes semantics and breaks the existing assumption (and fixtures) that GivTCP load is unreliable during some modes (e.g., grid charging). The repo’s REST fixtures show Load_Power != PV_Power + Grid_Power + Battery_Power (e.g., coverage/cases/rest_v2.json has Load_Power=624 but balance=590; rest_v3.json has 197 vs balance=233), and tests assert the energy-balance value. Consider reverting to the balance calculation, or make using Load_Power an opt-in config with a safe fallback when it disagrees significantly with the energy balance.

Suggested change

self.load_power = float(ppdetails.get("Load_Power", 0.0))

# Derive load power from energy balance rather than trusting REST Load_Power directly

self.load_power = self.pv_power + self.grid_power + self.battery_power

Copilot · 2026-02-19T08:54:49Z

templates/example_chart.yml

+      let res = []; for (const [key, value] of
+      Object.entries(entity.attributes.results)) { res.push([new
+      Date(key).getTime(), value]); } return res.sort((a, b) => { return a[0] -
+      b[0]  })


This data_generator assumes entity.attributes.results is always present; if the entity is unavailable or the attribute is missing, Object.entries(undefined) will throw and the chart will break. Consider using the same guard used later in this file (e.g., default to {} / return [] when results is falsy).

Suggested change

let res = []; for (const [key, value] of

Object.entries(entity.attributes.results)) { res.push([new

Date(key).getTime(), value]); } return res.sort((a, b) => { return a[0] -

b[0] })

if (!entity || !entity.attributes || !entity.attributes.results) return [];

const res = [];

for (const [key, value] of Object.entries(entity.attributes.results || {})) {

res.push([new Date(key).getTime(), value]);

}

return res.sort((a, b) => a[0] - b[0]);

Copilot · 2026-02-19T08:54:49Z

templates/example_chart.yml

+  - entity: sensor.predbat_load_ml_stats
+    attribute: load_today_h1
+    stroke_width: 2
+    color: green
+    curve: smooth
+    name: FC
+    opacity: 0.6
+    type: line
+    offset: "-1h"
+    extend_to: now
+    yaxis_id: forecasts
+  - entity: sensor.predbat_load_ml_stats
+    attribute: load_today_h8
+    stroke_width: 2
+    color: purple
+    curve: smooth
+    name: FC
+    opacity: 0.35
+    type: line
+    offset: "-8h"
+    extend_to: now
+    yaxis_id: forecasts


Legend/series names are duplicated (both the -1h and -8h series are named "FC"), which makes them indistinguishable in the legend/tooltip. Use distinct names (e.g., "FC -1h" and "FC -8h").

Copilot · 2026-02-19T08:54:49Z

templates/example_chart.yml

+  - entity: sensor.predbat_load_ml_stats
+    attribute: power_today_h1
+    stroke_width: 2
+    color: green
+    curve: smooth
+    name: FC
+    opacity: 0.6
+    type: line
+    offset: "-1h"
+    extend_to: now
+    yaxis_id: forecasts
+  - entity: sensor.predbat_load_ml_stats
+    attribute: power_today_h8
+    stroke_width: 2
+    color: purple
+    curve: smooth
+    name: FC
+    opacity: 0.35
+    type: line
+    offset: "-8h"
+    extend_to: now
+    yaxis_id: forecasts


Legend/series names are duplicated here as well (both forecast-offset series are named "FC"), which makes it hard to tell them apart in the legend/tooltip. Use distinct names for the -1h vs -8h series.

Copilot · 2026-02-19T08:54:49Z

templates/example_chart.yml

+      const res = [];
+      for (let i = 1; i < raw.length; i++) {
+        const [t1, e1] = raw[i-1];   // previous cumulative kWh
+        const [t2, e2] = raw[i];     // current cumulative kWh
+
+        const dt_hours = (t2 - t1) / 3600000;  // ms → hours
+        if (dt_hours <= 0) continue;
+
+        const dE = e2 - e1;  // kWh used during interval
+
+        const power_kw = dE / dt_hours;
+
+        res.push([t2, power_kw]);
+      }
+      return res;


This cumulative-kWh → power conversion doesn’t guard against negative deltas (e2 < e1), which can happen if the cumulative series resets/backsteps and will produce misleading negative power spikes. Consider skipping negative dE values (similar to the later rolling-average generator) or otherwise handling resets explicitly.

…the existing architecture if keeping the larger network is preferred. He initialisation for ReLU The weight initialisation currently uses Xavier/Glorot (std = sqrt(2/(fan_in + fan_out))), which was designed for linear/tanh activations. Since the network uses ReLU, switching to He initialisation (std = sqrt(2/fan_in)) would better maintain activation variance through the layers. #3388

springfall2008 added 2 commits February 18, 2026 15:28

New example ML chart

e69f775

Put back Load Power

04e5925

Copilot AI review requested due to automatic review settings February 19, 2026 08:50

Copilot started reviewing on behalf of springfall2008 February 19, 2026 08:51 View session

[pre-commit.ci lite] apply automatic fixes

d1e7ed7

Copilot AI reviewed Feb 19, 2026

View reviewed changes

springfall2008 added 2 commits February 19, 2026 09:08

Merge branch 'fixes20' of github.com:springfall2008/batpred into fixes20

8f6b706

springfall2008 changed the title ~~Example Load ML Chart, Put back GivTCP Load~~ Load ML Improvements, Put back GivTCP Load Feb 19, 2026

pre-commit-ci-lite bot and others added 5 commits February 19, 2026 09:12

[pre-commit.ci lite] apply automatic fixes

76dd55a

Fix failing tests

cdbb3b3

Merge branch 'fixes20' of github.com:springfall2008/batpred into fixes20

61d8b3b

Add MAE to docs

de534a0

Catch stop in octopus retry loop

c5b5408

springfall2008 merged commit 1867d8f into main Feb 19, 2026
1 check passed

springfall2008 deleted the fixes20 branch February 19, 2026 09:27

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Comments

Load ML Improvements, Put back GivTCP Load #3392

Load ML Improvements, Put back GivTCP Load #3392
springfall2008 merged 10 commits intomainfrom
fixes20

springfall2008 commented Feb 19, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Feb 19, 2026

Uh oh!

Copilot AI Feb 19, 2026

Uh oh!

Copilot AI Feb 19, 2026

Uh oh!

Copilot AI Feb 19, 2026

Uh oh!

Copilot AI Feb 19, 2026

Uh oh!

Copilot AI Feb 19, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	self.load_power = float(ppdetails.get("Load_Power", 0.0))
	# Derive load power from energy balance rather than trusting REST Load_Power directly
	self.load_power = self.pv_power + self.grid_power + self.battery_power

Uh oh!

Comments

Conversation

springfall2008 commented Feb 19, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 19, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants