Lynxby DOMINION
← Back to modules

EMS MODULE · ENERGY How much energy the plant should have consumed.

Comparing the kWh of two months says nothing if one of them made more, ran more shifts or was colder.

EMS is the Lynx energy management module. Consumption varies with output, with temperature and with shifts; EMS fits a baseline on those variables and contrasts actual consumption against it: the remaining difference is saving or deviation, not the effect of the period conditions.

Normalized baseline × actual metered consumption

Request a demo →

Add to the platform →

lynxplatform.cloud/ems/resumen

Product screens with sample data: no figure on these screens comes from a plant.

Six indicators, one decisive The one that does not go up just because you made more.

Every indicator is computed over the assets and the metering of the chosen scope, and carries its target, its direction —lower or higher is better— and its alarm threshold.

  • Total consumptionkWh

    Sum of consumption over the period for the assets and metering in scope.

  • Average consumptionkWh

    Mean consumption per observation, at the chosen analysis granularity.

  • Peak demandkW

    Highest power demanded in the period, with its timestamp.

  • Specific consumptionkWh/unit

    Consumption per unit produced. It is the indicator that does not vary with a change in volume.

  • Self-consumption%

    Share of consumption covered by on-site generation.

  • Custom formuladefined

    Your own KPI from the signals in scope, when none of the above fits.

Before the first number Three definitions before the first indicator

Comparing periods requires three prior definitions: what is measured, what explains that consumption, and the reference period the baseline is fitted on.

  • Scope and meteringThe assets and the consumption signal that enter the model, and the granularity they are aggregated to: hourly, daily, weekly or monthly.
  • Process variablesOutput, temperature, degree-days, shifts or any other signal that explains consumption. They are what normalizes the baseline: without them, the comparison reflects the change in conditions rather than the energy improvement.
  • Baseline and contrastThe model is fitted on the reference period. From then on, every new period is contrasted against the consumption the model predicts for those same conditions.

Variable analysis What explains consumption, before modelling it

A step before the baseline: you pick the target variable and every candidate, and the view computes which ones explain its variance. The impact of each one, the variance explained as they are added, and the correlation between them to drop the ones that contribute the same.

lynxplatform.cloud/ems/análisis-de-variables
Target variable (Y) · Electricity consumption (kWh)Target variance 80 %365 observations
Which variables weigh most?Bar = standardized coefficient β. ✓ = part of the recommended set.
  • Output (units)0.71
  • Compressors (kWh)0.44
  • Degree-days0.26
  • Active shifts0.12
  • Lighting (kWh)-0.05
How many do you need?Cumulative R² as variables are added by contribution.
52 %171 %284 %386 %487 %5
Do they overlap?
Output (units)
Compressors (kWh)
Degree-days
Active shifts
Lighting (kWh)
12345
Recommendation

With 3 variables you explain 84 % of the target: enough. Use them as the process variables of your EnPI baseline.

  • Impact comparable across scales

    The standardized coefficient β ranks variables measured in different units: output in units and temperature in degrees.

  • How many variables are needed

    The cumulative R² curve marks where the model stops improving. Adding variables past that point is overfitting.

  • Redundancies flagged

    Two correlated variables contribute almost the same. The matrix flags them and the view recommends keeping one.

Meanwhile, this month’s energy report still compares kWh against last year’s kWh. And the plant did not make the same amount.

A demo with your consumption →

Baseline A model that is chosen and justified

The wizard walks five steps: assets and metering, target variable and process variables, method, diagnostics and save. Methods are compared against each other with cross-validation, and the winner is the one that predicts best outside its own training period.

lynxplatform.cloud/ems/línea-base
  1. ✓Context
  2. ✓Variables
  3. ✓Method
  4. 4Diagnostics
  5. 5Save
Method
  • Driver auto-selection (CV)Recommended for your data
  • OLS · Linear regression
  • Segmented (production/standby)
  • Robust (Huber)
  • OLS + outlier filtering
  • Percentile by band (no drivers)
Compare models (ranking)
CV R²
Driver auto-selection (CV)0.82
OLS · Linear regression0.79
Segmented (production/standby)0.81
Robust (Huber)0.78
OLS + outlier filtering0.64
Diagnostics
  • Goodness of fitR² 0.84 · RMSE 214 kWh
  • Problematic observationsMax Cook 0.008 · threshold 0.011
  • Temporal validityDurbin-Watson 1.93 · CUSUM within band
  • Model assumptionsQ-Q on the diagonal
Reason for the operational changeReplacement of the two compressors in room 3Mandatory for ISO 50001 / 50006 traceability.
Save baseline
  • Six methods, one recommended

    OLS, segmented production/standby, Huber robust regression, OLS with outlier filtering, driver auto-selection by cross-validation, and percentile by time band for cases with no explanatory variables.

  • Diagnostics before saving

    Four questions with their charts and their metrics: how well it predicts (R², RMSE, MAPE), whether a single observation dominates the fit (Cook), whether the model still holds (Durbin-Watson, CUSUM) and whether its assumptions are met (Q-Q).

  • Traceability of the change

    Every baseline change asks for the operational reason —new line, shift change, equipment replacement— and is logged. That is what ISO 50001 and ISO 50006 ask for.

Comparison Actual versus expected consumption, with the model in sight

The period's consumption against what the baseline predicts, with the deviation as a percentage and the saving in energy units. The model's R² is shown next to the result: if the model is not reliable, neither is the comparison.

lynxplatform.cloud/ems/comparativa
  • Deviation and saving

    The difference between actual and expected, in kWh and as a percentage, over whichever period you choose.

  • Deviation alarm

    Every KPI carries a target, a direction and a threshold. When it is crossed, the platform warns without anyone having to open the screen.

  • When the baseline ages

    Residual CUSUM and Durbin-Watson detect drift: if consumption separates from the model in a sustained way, the baseline must be rebuilt.

EVIDENCE This is not a laboratory model.

A baseline needs history, and the history is already running at the installations published in the sectors section. Two glass melting furnaces at 15,000 kW with seven years measured. An iron foundry with ten furnaces whose first year on the platform is its baseline.

See the cases, sector by sector →

Lynx by Dominion

Your consumption against your own baseline

We prepare a demo with the consumption of one of your lines and the variables that explain it, and look at which baseline method fits your data.

Request a demo →

Or write to us directly at support.dt@dominion-global.com

← Back to modules