Machine learning can aid in making a model for inflow & infiltration
Sanitary sewer collection systems are expected to operate in perpetuity. The systems are in a constant state of renewal as various components reach their life expectancy at different times. In addition, some collection system components can decay faster than expected for a variety of reasons. Unmonitored and unmitigated decay in collection systems can lead to serious operational issues, such as sewer backups causing damage to private and public property, unlawful discharges into sensitive receiving waters, and risks to public health.
Inflow and infiltration occurs when there are cracks, defects or unintended connections into a sewer network. When one or more of these issues arises, water from external sources seeps into the sanitary system unintentionally, adding additional strain to the network. Inflow occurs when there is additional water flowing into the sanitary sewer system from direct points of entry, and infiltration from indirect points of entry via the surrounding soil.
During a significant rainfall event, this additional strain and added volume can exceed the sanitary sewer’s capacity, causing overflows of raw sewage at both inlets and outlets of the system. The additional strain from the increased volume also can lead to higher treatment costs, which are in turn passed along to the community. Monitoring and mitigating these overflows is critical when it comes to maintaining compliance and public safety.
Making a Model
Creating a model that provides an accurate analysis of inflow and infiltration (I&I) can be tedious and time-consuming. You need to go through the data and identify dry and wet weather patterns. Then, you should isolate rainfall events and corresponding sewer responses, subtract the dry weather flow from the wet-weather flow, then standardize the analysis through some form of statistical method to compare the collective response to a design criteria and level-of-service return period. This is where machine learning comes in handy.
Machine learning is the practice of applying statistical techniques to large amounts of data and looking for the best pattern to solve a problem. It then generates an implementation code that can recognize that pattern. This generated code is referred to as a model, and it can be called by applications that need to solve a particular problem. Machine learning in applications use this generated code to make better predictions. In other words, machine learning makes applications smarter, allowing the user to make better decisions. Machine learning has been used in many other industries, but has only been used to a limited degree in the municipal wastewater industry.
With enough data, a machine learning function, like the one built into FlowWorks, can identify wet and dry periods. The Machine Learning function automates the process and continuously updates the model as more data become available. Once calibrated, it also can help with quality control of new data by identifying what the flow should have looked like.
Using Machine Learning for I&I
1. Obtain flow and rainfall monitoring data
There are many different types of technology available to measure real-time flow, level and velocity in sewer systems. There also are many methods to analyze collected data. The basic concept is to derive base sanitary flow (domestic and institution/commercial/industrial flow), then determine the additional flow caused by I&I by subtracting the total flow from the base sanitary flow. The first step is to obtain a reasonable length of real-time rain and flow information (minimum 3 months, but optimally several years).
2. Define use case
The next step is to define a “use case” to test your model (model calibration and learning). This is essentially a test case that gives you an answer you already know. For instance, go back through your data and identify an I&I event. Then build a model with data pre-dating this event. You can include weather forecasts, infrastructure inventories, sensor data, etc., but you also must include the flow and rainfall data collected in Step 1. Then run the model as a verification scenario on a different event to see if you get a similar response. If successful, your model then is able to provide not only a forecast of the anticipated sanitary baseflow, but also a prediction of what the I&I could look like given a sufficient set of input data. The longer the dataset and the more information you train the model with, the better the prediction will be.
3. Run the statistical I&I package
Since all rainfall events are unique, best practices suggest that the I&I analysis should include a statistical assessment to derive return period flows such that basic service levels can be determined. This will allow I&I levels at 5- or 20-year levels to be determined. The statistics usually are derived from longer-term rainfall datasets where the statistics have already been run—for example, intensity, frequency, duration (IDF) curves. These curves can easily be entered into FlowWorks, resulting in a return-period flow to be derived from your data. This feature has always been available in FlowWorks, but now it includes automatic storm selection.
4. Compare to original design criteria
Understanding the level of I&I within the sewer system as the infrastructure ages and comparing to original design criteria is a necessary best practice for infrastructure management. Now that you have the return period derived flow, you can compare back to what the original design called for and understand if there is a reduction in level of service to your customers. Performing this task at 5- and 10-year intervals and comparing the results also helps to understand the state of the infrastructure and how gracefully it is aging.
5. View the results
Using Machine learning saves time and increases productivity. No need to go through defining dry weather and wet weather patterns, which can take a long time. Using an application like FlowWorks, this process is now automated. The longer you are collecting data, the better and more accurate the model becomes.
Regardless of which application you choose to use, introducing machine learning can dramatically simplify what was previously an extremely tedious task. Automation allows you to monitor your city infrastructure and water management systems more efficiently to protect your sewage and storm water infrastructure.