The Implementation of Artificial Intelligence for Optimizing Odor Control Chemical Dose Rates

Problem: Hydrogen sulfide gas (H2S) is a foul smelling, toxic, and corrosive gas that is generated by natural processes through the normal operation of wastewater conveyance and treatment systems. While odor abatement can be obtained through both chemical and mechanical means, the majority of the market continues to rely on chemicals metered into the wastewater system to capture, oxidize, or otherwise prevent the formation of H2S. This 'liquid phase' treatment presents the unique problem of ever-changing demand as wastewater system parameters change from minute to minute. While systems can be manually optimized by evaluating how conditions are changing and adjusting chemical dose rates to accommodate, this is a time-consuming effort, that relies on the skill of the individual, and is typically performed no more than twice per year during seasonal changes. Furthermore, the process is not repeatable across a large number of odor control sites with multiple practitioners as the individual sentiments of each individual come into play when optimizing against two inversely proportional variable, for example H2S level and budget. Objective: In the proposed paper, we will demonstrate the novel use of Artificial Intelligence to optimize odor control chemical dosing rates on a weekly basis. This paper will show that through the use of Artificial Intelligence, the practitioner was able to achieve as good or better H2S control using industry standard measurement methods while maintaining or reducing the total volume of chemical required.
Theoretical Methodology and Analysis: The AI platform leverages new and historic data including wastewater temperature, flow rates, hydrogen sulfide levels, and past dosing history as well as immutable characteristics of the collection system such as pipe length and diameter to make modifications to pre-existing chemical dose curve ml/min setpoints, or ppm setpoints if the chemical dose is designed to change proportionally to the wastewater flow rate. The AI model produces a week of H2S predictions into the future. Each hour dosing setpoint is evaluated to see if the given dose will result in a predicted H2S concentration above or below the H2S control point, the model then tries to derive a dosing setpoint that brings the H2S levels below the threshold up, and H2S levels above the threshold down. This process is then repeated until the budget is met and the optimal dosing curve is returned.
Core to this process are two dimensions that the process is optimized against, H2S concentration and chemical volume. H2S is optimized against static upper and lower control limits that act as boundaries inside of which is considered an acceptable H2S measurement. The second dimension of chemical volume is established based on the specific goals of the program, to either minimize the use of chemical or ensure that a certain volume of chemical is used in a specified period of time. Application: The chemical dosing AI was applied at chemical odor control sites in two separate east coast utilities, one in the northern and one in the southern United States, to evaluate its effectiveness at continuously optimizing odor control chemical dose rates. To calculate the amount of product required to treat the considered sewer segment, the following parameters were employed: wastewater flow and temperature, sewer diameter and length, historic odor control chemical feed rates, historic atmospheric H2S data with and without odor control chemical feed and the corresponding retention time calculation.
Case 1: A wastewater pump station (PS 6) in Monroe Township, New Jersey has been optimized with the help of the AI for the past 11 months (from January 2021 till present) to maintain the H2S levels at an average of 10 ppm and a peak of 20 ppm. Based on the historic data along with the past week's data, the AI tool predicts the H2S levels of the upcoming week and delivers a corresponding dose curve. This dose curve will be updated on the advanced chemical dose controller weekly to feed the product into the sewer segment in real time. The overall performance of this site was measured by comparing the dosage over the last 11 months (Figure 1) along with the corresponding average H2S levels (Figure 2) on a month-by-month basis for 2020 vs 2021. The data indicates that H2S concentrations were maintained well within the target limits and at a lower average level throughout the period of monitoring, without increasing the required chemical volume. Furthermore, the month-over-month comparison of the H2S and feed data shows a constant improvement in process control to keep the concentrations around 10 ppm.
Case 2: A coastal utility in Southeast United States has been optimized with the help of the AI over the past couple of months since October 2021 until present to maintain the H2S levels at an average of 50 ppm and a peak of 150 ppm. In Figure 3 and Figure 4, the H2S concentration and the product feed data are compared for the months of October and November in 2020 vs 2021. The chemical dosing AI, was able to optimize the chemical dosing to meet the demand, resulting in a 72.7% reduction in H2S for the month of October and a 75% reduction in measured H2S in November without increasing the volume of chemical use. The H2S emissions were targeted precisely with an advanced curve and a more continual monitoring than the conventional optimization process. In both the cases discussed above the chemical dosing AI was able to maintain a lower H2S concentration at an equal or lower chemical volume requirement than during the same period the previous year. Above all, the chemical dosing AI was able to effectively optimize odor control chemical dosing rates more effectively than manual optimization, reducing the level of human effort required considerably.