⚡ electrical engineering

Regression-Based Prediction of Phosphate Concentration in Drum Water Systems under Ammonia–Phosphate Treatment in Power Plants

This study presents a highly accurate single-variable regression model based on the natural logarithm of electrical conductivity to predict phosphate concentrations in ammonia–phosphate treated drum water systems, offering a simplified and reliable tool for real-time boiler water quality management without the need for stream-specific adjustments.

Original authors: Ahmad Zamani, Mortezs Fazeli, Mohamad Khatami

Published 2026-06-30

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Ahmad Zamani, Mortezs Fazeli, Mohamad Khatami

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a power plant as a giant, high-pressure tea kettle. Inside this kettle, water is boiled to create steam that spins turbines and generates electricity. But just like a regular kettle can get clogged with limescale or rusted from the inside, these massive industrial boilers are at risk of corrosion and buildup if the water chemistry isn't perfect.

To keep the water healthy, plant operators add a special "vitamin mix" called ammonia-phosphate treatment. The phosphate is the star player here; it acts like a protective shield, preventing rust and keeping the water stable. However, there's a catch: you need to know exactly how much phosphate is in the water at all times. Too little, and the metal rusts; too much, and it can cause other problems.

The Old Way: The Slow Lab Test

Traditionally, checking the phosphate level is like sending a sample of your soup to a fancy restaurant kitchen to be tasted by a chef. It's accurate, but it takes time. You have to grab a cup of water, send it to a lab, mix it with chemicals, wait for the color to change, and then get a result. By the time you get the answer, the water chemistry in the boiler might have already changed. It's too slow for real-time control.

The New Idea: The "Sound Check"

The researchers in this paper asked a simple question: Can we guess the amount of phosphate just by listening to how the water "conducts" electricity?

Think of electrical conductivity like the volume of a radio.

If the water is pure, the "volume" (conductivity) is low.
If you add salts or chemicals (like phosphate), the "volume" goes up because there are more particles carrying the signal.

The team collected 433 samples of water from a real power plant in Iran. They measured two things for each sample:

The actual phosphate level (the "real" answer).
The electrical conductivity (the "volume" of the water).

The Discovery: A Simple Math Trick

They found that the relationship wasn't a straight line (like a ladder), but more like a curve. To fix this, they used a mathematical trick called a natural logarithm (think of it as a special lens that straightens out a curved road).

When they looked at the data through this "logarithmic lens," they found a perfect match. They created a simple formula:

Phosphate Level = (5.6 × Log of Conductivity) – 12.3

This formula is like a magic decoder ring. Instead of waiting for a lab test, an operator can look at the conductivity meter (which is always on and always working) and instantly know the phosphate level.

How Well Does It Work?

The results were impressive. The model predicted the phosphate levels with 89% accuracy (a score of 0.89 out of 1.0).

The Error Margin: On average, the prediction was off by less than 1 part per million. In the world of power plant chemistry, that is incredibly precise.
The Benefit: It works for all the samples they tested without needing to be tweaked for different parts of the system. It's a "one-size-fits-all" rule for this specific plant.

The Catch (What the Paper Actually Says)

The authors are very honest about the limitations. This "magic decoder ring" was built using data from one specific power plant with a specific type of water treatment.

It's like a recipe that works perfectly for your kitchen with your stove. If you take that same recipe to a different house with a different stove, you might need to adjust the heat.
The paper states that if you try to use this exact formula in a different power plant with different water conditions, you might need to "recalibrate" it (tweak the numbers) to make it work again.

The Bottom Line

This study shows that you don't always need a slow, complex lab test to monitor boiler water. By simply measuring how well the water conducts electricity and plugging that number into a simple formula, power plant operators can get a fast, reliable estimate of phosphate levels. This helps them keep the boiler safe, prevent rust, and keep the electricity flowing without the delay of waiting for lab results.

Technical Summary: Regression-Based Prediction of Phosphate Concentration in Drum Water Systems

Problem Statement
In modern steam-generating power plants, maintaining precise boiler water chemistry is critical for preventing corrosion, scale formation, and operational inefficiencies. Specifically, under ammonia–phosphate treatment regimes, accurate monitoring of phosphate concentration is essential to avoid issues such as phosphate hideout, deposition, and carryover. While conventional wet-chemical methods (e.g., the molybdate method per ASTM D515) provide accurate measurements, they are time-consuming and unsuitable for continuous, real-time online monitoring. Existing indirect measurement approaches often rely on multi-parameter models or require stream-specific calibrations, which can increase complexity and reduce robustness when operational conditions fluctuate. There is a practical need for a simplified, robust predictive model capable of estimating phosphate concentration directly from easily measurable parameters without extensive recalibration.

Methodology
The study utilized a dataset of 433 operational water samples collected from high-pressure (HP) and low-pressure (LP) drum water systems at the Ferdowsi Combined Cycle Power Plant in Iran. The plant operates under ammonia–phosphate treatment conditions.

Data Collection: Samples were collected in 500 mL containers during standard shifts at 25°C.
Measurements: Electrical conductivity (SC) was measured online using Hamilton H100 and Swan AMI analyzers. Phosphate concentrations were determined via the molybdate method (ASTM D515). Ammonia concentration was initially measured (ASTM D1426) but excluded from the final model due to weak correlation with phosphate levels compared to conductivity.
Model Development: An exploratory analysis identified the natural logarithm of specific conductivity, $\ln(SC)$ , as the strongest predictor. A single-variable linear regression model was developed using Microsoft Excel and the least-squares fitting method.
Evaluation: The model's performance was assessed using the coefficient of determination ( $R^2$ ), root mean square error (RMSE), and mean absolute error (MAE).

Key Contributions
The primary contribution of this work is the development of a simplified, single-variable regression model that predicts phosphate concentration solely based on the natural logarithm of specific conductivity. The resulting equation is:
$PO_4 = 5.6\ln(SC) - 12.3$
Where $PO_4$ is the phosphate concentration in ppm and $SC$ is the specific conductivity in $\mu S/cm$ .

Unlike traditional empirical methods that may require complex multi-parameter inputs or frequent stream-specific adjustments, this approach demonstrates that a single, continuously available parameter (conductivity) can effectively estimate phosphate levels. The use of the logarithmic transformation of conductivity was found to improve linearity and reduce the influence of extreme observations compared to direct conductivity fitting.

Results
The model demonstrated strong predictive accuracy across the 433 samples:

Coefficient of Determination ( $R^2$ ): 0.89, indicating that a substantial proportion of the variability in phosphate concentration is explained by conductivity measurements.
Root Mean Square Error (RMSE): 0.76 ppm.
Mean Absolute Error (MAE): 0.55 ppm.

The results confirm a strong relationship between phosphate concentration and $\ln(SC)$ , validating the hypothesis that phosphate species significantly contribute to the ionic strength of boiler water under ammonia–phosphate treatment conditions.

Significance and Claims
The authors position this model as a practical operational tool rather than a universal correlation. Its significance lies in offering a reliable method for real-time monitoring and control of boiler water chemistry, which supports the prevention of corrosion and scaling. By reducing reliance on time-consuming laboratory analyses, the model facilitates more efficient cycle chemistry surveillance and early detection of deviations from target concentrations.

However, the paper maintains a modest scope regarding its applicability. The authors explicitly state that the dataset was derived from a single power plant under specific treatment conditions. Consequently, they caution that recalibration may be necessary before applying the model to systems with different water chemistry characteristics or varying sodium-to-phosphate ratios. The study concludes that while the model is a robust strategy for continuous water quality management in similar industrial boiler systems, future work should expand the dataset to multiple power plants to further improve general applicability.