Cross-checking extrusion energy savings with intensity and regression baselines

01 · CASE ANALYSIS

Separate production change from efficiency

Reference LINE A retained its operating conditions while LINE B received improvements. Average electricity fell, but average production weight also fell, making raw electricity reduction an unsuitable savings metric.

The intensity comparison restricts the reference production range for comparability. Customer, factory and actual line identifiers are anonymized.

02 · CASE ANALYSIS

Method 1: production-normalized intensity

Average individual intensity changed from 0.234932 to 0.202304 kWh/kg, a reported 13.9% improvement. Mean electricity changed from 147.70 to 119.26 kWh and mean production from 627.45 to 596.21 kg.

The ratio of average electricity to average production gives a weighted intensity improvement of approximately 15.0%. This differs from averaging individual intensities and should be labeled separately.

03 · CASE ANALYSIS

Method 2: adjusted regression baseline

After comparing OLS, WLS and GLSAR and reviewing influential observations, the source selected GLSAR with AR(1) errors. The reported equation is E = −26.2148 + 0.8283 × time-utilization (%) + 0.1612 × production weight (kg).

Reported diagnostics were adjusted R² 0.801, RMSE 7.64 kWh, CV(RMSE) 4.73%, VIF 1.089 / 1.089 and Durbin–Watson 1.958. These describe the model, not a savings guarantee.

04 · CASE ANALYSIS

Report the valid subset and coverage

Of 50 reporting observations, 28 were valid (56.0%) and 22 invalid (44.0%). Valid adjusted-baseline electricity was 4,366.59 kWh against actual 3,683.00 kWh: 683.59 kWh, or 15.65%.

Reference production range was 493.95–881.10 kg and utilization 80.00–95.45%; reporting ranges were 80.10–747.60 kg and 65.15–95.83%. The original validity rule also used actual electricity, which can influence selection. For reuse, define eligibility from independent operating variables before evaluating savings.

05 · CASE ANALYSIS

Use both methods without combining results

The methods support the same improvement direction but apply different normalization and selection. Do not add 13.9% and 15.65%, or extrapolate the valid-subset kWh to the whole factory.

Collect broader baseline coverage, product mix, quality, downtime and synchronized metering for subsequent evaluation. Report model uncertainty and excluded observations alongside savings.

Source-value comparison

Source figures · English edition

Figures correspond to those published in the Korean case. English titles, axes and legends are provided; published values and evidence boundaries are retained. Open a figure at full size to inspect small labels.