{"id":{"repo_id":"csusb","oai_identifier":"oai:scholarworks.lib.csusb.edu:etd-3450"},"canonical_url":"https://search.ndltd.org/etd/csusb/oai:scholarworks.lib.csusb.edu:etd-3450","repository":{"repo_id":"csusb","name":"CSUniversity San Bernardino","base_url":"https://scholarworks.lib.csusb.edu/do/oai/"},"display":{"title":"TIME SERIES DEEP LEARNING APPROACH FOR THE INTERMITTENT OPERATIONAL PERFORMANCE OF A WELLHEAD WATER TREATMENT AND DESALINATION SYSTEM","abstract":"Distributed water treatment and desalination (DWTD) systems are becoming significant for serving disadvantaged communities that are geographically segregated from centralized water distribution networks. However, given the remote nature of the communities, these systems must operate autonomously adapting to intermittent operations due to varying water use patterns and unavailability of continuous manual labor support. Machine Learning models describing and forecasting system performance are critical, allowing for model-based control, performance forecasting, fault detection, and determination of causal relationships among process attributes. Accordingly, graph convolutional neural networks with an attention mechanism (GATConv) were developed to describe the intermittent operational profiles of a wellhead water treatment system deployed in a small remote disadvantaged community in California. Time-series data from 22 system sensors was compiled for the four operational modes (startup, permeate production, shutdown, flushing) and three outcomes (nitrate passage (%), salt passage (%), and permeate flux (LMH)). GATConv models with 3 hidden layers, 16 hidden nodes and 8 attention heads in each layer were developed based on 4 months of operational data, consisting of over 6 million samples, each containing 22 sensor data points. GATConv models demonstrated excellent accuracy with R2 > 0.95 for prediction of permeate flux for up to 1 month of operational data forward in time. Causal relationships were extracted by aggregating and visualizing the learned weights within the network structure. For production mode, connections from {feed flow rate to RO system, permeate flow rate, flow rate feed to pressure tank, permeate temperature} were with higher weight for permeate flux, and {raw well feedwater flow rate, recycle stream flow rate, inlet feed pressure for RO elements} showed high weights for feed flow rate. For nitrate passage, permeate conductivity, recycle stream flow rate, raw well feedwater flow rate, and inlet feed pressure for RO element were with higher weight to feed flow rate. Analysis of significant model attributes (normalized weights accumulated over GATConv layers) demonstrated higher weights for {inlet feed pressure, feed flow rate, permeate temperature, permeate flow rate, concentrate pressure}.","abstract_html":"Distributed water treatment and desalination (DWTD) systems are becoming significant for serving disadvantaged communities that are geographically segregated from centralized water distribution networks. However, given the remote nature of the communities, these systems must operate autonomously adapting to intermittent operations due to varying water use patterns and unavailability of continuous manual labor support. Machine Learning models describing and forecasting system performance are critical, allowing for model-based control, performance forecasting, fault detection, and determination of causal relationships among process attributes. Accordingly, graph convolutional neural networks with an attention mechanism (GATConv) were developed to describe the intermittent operational profiles of a wellhead water treatment system deployed in a small remote disadvantaged community in California. Time-series data from 22 system sensors was compiled for the four operational modes (startup, permeate production, shutdown, flushing) and three outcomes (nitrate passage (%), salt passage (%), and permeate flux (LMH)). GATConv models with 3 hidden layers, 16 hidden nodes and 8 attention heads in each layer were developed based on 4 months of operational data, consisting of over 6 million samples, each containing 22 sensor data points. GATConv models demonstrated excellent accuracy with R2 &gt; 0.95 for prediction of permeate flux for up to 1 month of operational data forward in time. Causal relationships were extracted by aggregating and visualizing the learned weights within the network structure. For production mode, connections from {feed flow rate to RO system, permeate flow rate, flow rate feed to pressure tank, permeate temperature} were with higher weight for permeate flux, and {raw well feedwater flow rate, recycle stream flow rate, inlet feed pressure for RO elements} showed high weights for feed flow rate. For nitrate passage, permeate conductivity, recycle stream flow rate, raw well feedwater flow rate, and inlet feed pressure for RO element were with higher weight to feed flow rate. Analysis of significant model attributes (normalized weights accumulated over GATConv layers) demonstrated higher weights for {inlet feed pressure, feed flow rate, permeate temperature, permeate flow rate, concentrate pressure}.","abstract_has_math":false,"creators":["Clement, Michael G"],"institution":null,"degree_name":"Master of Science in Computer Science","degree_level":"Project","degree_discipline":"School of Computer Science and Engineering","degree_department":null,"school":null,"contributors":["Bilal Khan"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":null,"date_issued":"","date_published":null,"updated_at":"2026-11-19T08:10:04Z","subjects":["water treatment","deep learning","artificial intelligence","desalination","Artificial Intelligence and Robotics","Computer Sciences"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://scholarworks.lib.csusb.edu/etd/2244","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://scholarworks.lib.csusb.edu/etd/2244","record_label":"Repository record","record_kind":"record","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://scholarworks.lib.csusb.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Ascholarworks.lib.csusb.edu%3Aetd-3450","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Bilal Khan"]},{"key":"dc:creator","label":"Author","values":["Clement, Michael G"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.available","label":"Dc Date Available","values":["2025-04-14T07:00:00Z"]},{"key":"dc:date.created","label":"Dc Date Created","values":["2025-05-01T07:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["School of Computer Science and Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Project"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science in Computer Science"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["water treatment","deep learning","artificial intelligence","desalination","Artificial Intelligence and Robotics","Computer Sciences"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://scholarworks.lib.csusb.edu/etd/2244"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Distributed water treatment and desalination (DWTD) systems are becoming significant for serving disadvantaged communities that are geographically segregated from centralized water distribution networks. However, given the remote nature of the communities, these systems must operate autonomously adapting to intermittent operations due to varying water use patterns and unavailability of continuous manual labor support. Machine Learning models describing and forecasting system performance are critical, allowing for model-based control, performance forecasting, fault detection, and determination of causal relationships among process attributes. Accordingly, graph convolutional neural networks with an attention mechanism (GATConv) were developed to describe the intermittent operational profiles of a wellhead water treatment system deployed in a small remote disadvantaged community in California. Time-series data from 22 system sensors was compiled for the four operational modes (startup, permeate production, shutdown, flushing) and three outcomes (nitrate passage (%), salt passage (%), and permeate flux (LMH)). GATConv models with 3 hidden layers, 16 hidden nodes and 8 attention heads in each layer were developed based on 4 months of operational data, consisting of over 6 million samples, each containing 22 sensor data points. GATConv models demonstrated excellent accuracy with R2 > 0.95 for prediction of permeate flux for up to 1 month of operational data forward in time. Causal relationships were extracted by aggregating and visualizing the learned weights within the network structure. For production mode, connections from {feed flow rate to RO system, permeate flow rate, flow rate feed to pressure tank, permeate temperature} were with higher weight for permeate flux, and {raw well feedwater flow rate, recycle stream flow rate, inlet feed pressure for RO elements} showed high weights for feed flow rate. For nitrate passage, permeate conductivity, recycle stream flow rate, raw well feedwater flow rate, and inlet feed pressure for RO element were with higher weight to feed flow rate. Analysis of significant model attributes (normalized weights accumulated over GATConv layers) demonstrated higher weights for {inlet feed pressure, feed flow rate, permeate temperature, permeate flow rate, concentrate pressure}."]},{"key":"dc:title","label":"Title","values":["TIME SERIES DEEP LEARNING APPROACH FOR THE INTERMITTENT OPERATIONAL PERFORMANCE OF A WELLHEAD WATER TREATMENT AND DESALINATION SYSTEM"]}]}],"canonical_facts":{"dc:contributor":["Bilal Khan"],"dc:creator":["Clement, Michael G"],"dc:date.available":["2025-04-14T07:00:00Z"],"dc:date.created":["2025-05-01T07:00:00Z"],"dc:description.abstract":["Distributed water treatment and desalination (DWTD) systems are becoming significant for serving disadvantaged communities that are geographically segregated from centralized water distribution networks. However, given the remote nature of the communities, these systems must operate autonomously adapting to intermittent operations due to varying water use patterns and unavailability of continuous manual labor support. Machine Learning models describing and forecasting system performance are critical, allowing for model-based control, performance forecasting, fault detection, and determination of causal relationships among process attributes. Accordingly, graph convolutional neural networks with an attention mechanism (GATConv) were developed to describe the intermittent operational profiles of a wellhead water treatment system deployed in a small remote disadvantaged community in California. Time-series data from 22 system sensors was compiled for the four operational modes (startup, permeate production, shutdown, flushing) and three outcomes (nitrate passage (%), salt passage (%), and permeate flux (LMH)). GATConv models with 3 hidden layers, 16 hidden nodes and 8 attention heads in each layer were developed based on 4 months of operational data, consisting of over 6 million samples, each containing 22 sensor data points. GATConv models demonstrated excellent accuracy with R2 > 0.95 for prediction of permeate flux for up to 1 month of operational data forward in time. Causal relationships were extracted by aggregating and visualizing the learned weights within the network structure. For production mode, connections from {feed flow rate to RO system, permeate flow rate, flow rate feed to pressure tank, permeate temperature} were with higher weight for permeate flux, and {raw well feedwater flow rate, recycle stream flow rate, inlet feed pressure for RO elements} showed high weights for feed flow rate. For nitrate passage, permeate conductivity, recycle stream flow rate, raw well feedwater flow rate, and inlet feed pressure for RO element were with higher weight to feed flow rate. Analysis of significant model attributes (normalized weights accumulated over GATConv layers) demonstrated higher weights for {inlet feed pressure, feed flow rate, permeate temperature, permeate flow rate, concentrate pressure}."],"dc:identifier":["https://scholarworks.lib.csusb.edu/etd/2244"],"dc:subject":["water treatment","deep learning","artificial intelligence","desalination","Artificial Intelligence and Robotics","Computer Sciences"],"dc:title":["TIME SERIES DEEP LEARNING APPROACH FOR THE INTERMITTENT OPERATIONAL PERFORMANCE OF A WELLHEAD WATER TREATMENT AND DESALINATION SYSTEM"],"thesis:degree_discipline":["School of Computer Science and Engineering"],"thesis:degree_level":["Project"],"thesis:degree_name":["Master of Science in Computer Science"]},"updated_at":"2026-11-19T08:10:04Z"}