PREDICTIVE AND OPERATIONAL ANALYTICS, WHAT IS IT REALLY ALL ABOUT? Derek Vogelsang 1, Alana Duncker 1, Steve McMichael 2 1. MWH Global, Adelaide, SA 2. South Australia Water Corporation, Adelaide, SA ABSTRACT In a world of constrained water resources and extreme weather events that can often disrupt water supply, effective water management is paramount. The challenges of managing water supply and security have contributed to rising costs for customers in recent years. Water utilities across the board are reducing costs of supply and to looking at ways to deliver water more efficiently. Specific technologies can enable infrastructure to be managed more efficiently, and drive savings for the end user. The interconnectivity works undertaken for the metropolitan network included new transfer pipelines, new pump stations and new pressure reducing valves to change the way water is transferred to or through existing pressure zones. These upgrades to the network are shown in Figure 1 and give SA Water to the flexibility to operate in a number of different transfer modes and ensure that most customers in the metropolitan network can be supplied from at least two water sources (water treatment plants including the desalination plant). INTRODUCTION At its core, predictive and operational analytics is about analysing current and historical data to make predictions about the future. This paper presents a real world example of predictive and operational analytics applied to the water industry. Water companies are looking for ways to reduce operating cost whilst improving operational efficiency and reliably of their networks. The ability to predict the performance a network and forecast conditions (such as demands) are essential to planning and delivering operational efficiency. An integrated suite of decision support tools have been developed for South Australia Water Corporation to deliver this capability for their large and complex Adelaide Metropolitan water distribution network. BACKGROUND South Australia Water Corporation (SA Water) has recently completed a $400 million investment in the Adelaide metropolitan water distribution network to improve flexibility and reliability of supply, improve water security during prolonged drought and increase capacity in the network to allow for growth until 2050. This project was known as the North South Interconnection System Project (NSISP) and was a key part of South Australia s Network Water Security Program, aimed at achieving long term water security for Adelaide and optimal efficiency in its use and management. Figure 1: Infrastructure Upgrades to the Network As part of this project SA Water required decision support tools to help the business and operators manage their network and allow the business to make the most of the new flexibility enabled through the interconnection works. MWH Global as part of the Waterlink Joint Venture, which delivered the North South Interconnection System Project (NSISP), developed a suite of
sophisticated and ground breaking predictive and operational analytics tools for SA Water. Four Decision Support Tools were identified through a process of user workshops, high level business requirements and functional requirements, these were: Demand Forecast Tool Distribution Optimisation Tool Network Operations Model Network Status Display Although each of the tools has a separate function and use different technologies, close integration of the tools was a requirement. The integration and interdependencies of the tools is shown in Figure 2. Figure 2: Decision Support Tools for SA WATER OVERVIEW OF THE TOOLS Demand Forecast Tool (DFT) The Demand Forecast Tool uses a regression model to develop a relationship between demand and a number of climate variables (temperature, rainfall, evaporation and soil moisture). The regression model was built using historic weather, customer billing (demand) and water treatment plant production data. Each day the tool receives a 7 day weather forecast from the Bureau of Metrology and using the regression model calculates a demand forecast for the network dependent on the climate forecast. The tool then carries out additional calculations to break this down to the customer level and develops diurnal patterns for each customer type in 30 minute increments for the next 7 days. This forecast is then used in other predictive models such as the Network Operations Model. Figure 3 illustrates the concept design for the DFT showing the data inputs, high level calculation process and then outputs to other tools. Figure 3: DFT Concept Design The Demand Forecast Tool allows SA Water to forecast the water treatment plant production needs of their metropolitan network and also plan around potential production restrictions. This is possible as DFT spatially distributes the whole of system forecast demand by firstly breaking down demand per customer type and then per customer using billing data collected from the SA Water Customer data base. Demand forecasts per customer can then be allocated to zones (or demand areas ) and further allocated to treatment plant areas to determine production needs for the next 7 days. Calculating diurnal curves for each customer type depending on the 7 day demand forecast, allows the Network Operations Model to make use of this forecast as well. Keeping the Network Operations Model (NOM) up to date with the latest demands means modelling of network performance under normal and un-planned incidents can be based on the best possible demand inputs available. Figure 4 shows the Demand Forecast vs Actual Production and also the breakdown to diurnal patterns per customer type. Figure 4: Demand Forecast Tool
Distribution Optimisation Tool (DOT) The Distribution Optimisation Tool takes a whole of system approach to optimise water source selection, water treatment plant production forecasts and metropolitan network configuration (operations planning). The optimisation tool uses current reservoir storage levels from SCADA and forecast demands to make decisions on how best to configure and run the Adelaide water system in the short (28 days),and long term (2 years). ensuring the balance of inputs and outputs in the network and changes in storage levels over time in the solutions. Figure 5 shows the model schematic. This decision support tool allows the business to identify operational efficiencies and plan their whole of system operations based on the current and forecast conditions. The tool uses a linear programming based optimisation algorithm to solve for minimum cost. A linear program requires an objective function to minimise or maximise and constraint or rules to guide the optimisation. In the DOT, the objective function takes into account major pumping costs in the bulk water network, water treatment plant costs and metropolitan transfer pumping costs. Natural inflows to the system, available storage volumes, demands on the network (metro and country) as well as constraints like minimum and maximum flows in pipelines and rivers, and minimum or maximum storage volumes form part of the network constraints formulation. This whole of system Source to Tap approach makes sure all aspects of the water balance are taken into account. A high-level water balance model describes the interconnectivity of the system, the allowable flow paths, the options for storage and the source of water. The linear program optimises the flows in each of the links of the network model over time. The mass balance model is closely linked to the optimisation routine and is used in developing the network constraints for the linear program, and Figure 5: Distribution Optimisation Tool (schematic) Network Operations Model (NOM) The Network Operations Model (NOM) has been built using Innovyze s InfoWorks and IWLive products. As an all mains hydraulic model, the NOM contains approximately 100,000 pipes and nodes as well as 95 tanks, 110 pumps and 270 control valves (See Figure 6). As a live model, the NOM is scheduled to run a 7 day simulation every hour and uses current boundary conditions (live data from SCADA for tank levels, pump on/off and valve on/off status) and forecast demands (per customer category from the DFT) in each simulation. This gives the business users a current model to review predicted network performance and any highlighted upcoming issues. It also means a current model is available to run incident and response scenarios as they happen to give better hydraulic information to field staff and understanding of network performance. Figure 6: Network Operations Model (in IWLive)
Typically the purpose of a hydraulic model is to predict the performance of a water network (i.e flows in pipes, pressures as nodes and levels in storages) for design or planning purposes. Under a peak day or future demand conditions, the model will calculate flows and pressures and users can view and investigate the performance of the assets and determine whether low pressures or water shortages are likely and if additional infrastructure is required. However, hydraulic models can also be used as operational models, modelling conditions in the network as they currently are or are predicted to be in the short term. This allows users to predict upcoming network issues, review and model operational changes, model planned and unplanned maintenance issues or outages and plan operational responses for field staff to implement. The Network Operations Model (NOM) built for SA Water is an operations model, with users including network controllers, production planners, modellers and engineers who work in SA Water s the Operations Control Centre. The NOM is a Live hydraulic model, meaning that the model is always up to data with live data from the field setting the starting conditions and data from the latest demand forecast (up to 400 live data feeds in this case). Having a live model saves time during incident modelling (i.e pipe burst) as the boundary conditions and demand data sets do not have to be configured. Alarms and warnings have also been setup for the NOM so that potential issues in hydraulic performance are flagged to the users ahead of time (i.e future low pressure in the network due to increased demand). Live models such as this do need maintenance from experience modellers to ensure latest controls (i.e tank/valve/pump set points) are in line with settings in the field, model calibration tasks are carried out, and customer base demands are in-line with the seasonal customer billing data used in the Demand Forecast Tool. In the case of the NOM for SA Water, the experienced modellers in the control room are responsible for the maintenance of the model and also setting up the Baseline Model through Innovyze s the InfoWorks software. IWLive sits alongside InfoWorks as the access point to the Live hydraulic models. IWLive runs the baseline setup by the modellers and runs on an automatic scheduled so that latest hydraulic results, alarms and warnings are always available to the controllers and other users. All incident and response modelling by the control room operators is also done through IWLive. Network Status Display (NSD) The Network Status Display is a web-based dashboard used to bring together actual and historical data from a range of business systems (GIS, SCADA, Water Quality Samples, Customer Billing) and present the results of the predictive models (demand forecasting and hydraulic model performance from the NOM). The Network Status Display (NSD) allows users to interpret historic data, get easy access to data, visualise predicted performance data, compare actual and predicted performance, generate reports and make decisions. Figure 7 shows the Network Status Display on the Network Status page. Figure 7: Network Status Display
The Network Status Display (NSD) Tool was designed as a means for providing a more intuitive and information-rich display of network status than is currently provided to operations controllers through SCADA-based displays. Key to the concept is utilisation of the GIS environment to provide realistic spatial depictions of the network and to easily overlay data and information from a diverse range of sources depending on the user and situation. A further major role of the system is to make such data accessible to a wider range of SA Water business users and to enable user-specific interfaces and data access to be readily provided. The tool graphically displays the current network status (water treatment plant service areas) and utilises a geographic information system interface to show details such as pressure zones, street names, land parcels, water quality sample points, pipelines, tanks and other assets included in the hydraulic model. Facility information (from GIS) Data from other decision support tools: Demand Forecast Tool Results Forecast production rates Forecast tank, pump, valve performance from the Network Operations Model (NOM) Forecast pressures for major customers (NOM) Figure 8 shows calculated production planning data, live tank level and flow meter data, water quality data and forecasted tank level results from the Network Operations Model. The Network Status Display has been built using C3 Global s Amulet product. Amulet acts as a portal into many different data sources and displays the information in a web based format in a variety of charts, tables and graphics. Amulet can also perform complex calculations on the data it collects and display the results through the dashboards. For the Network Status Display, the project team configured Amulet to display different types or groups of data across a number of Tabs. This data can also be filtered via a Network Navigation Tree along the site (collecting data by Zone, Facility, Area, etc). Display widgets in the dashboard have been configured to display the information relevant to SA Water s business users, such as: Actual and historic performance data from SCADA: Flow meter data for production calculations Tank and reservoir level data Pump station flows and on/off status Valve station on/off status On-line water quality readings Data from other business systems: Water quality sample point data Actual customer billing data Actual and forecast weather from BOM Figure 8: Water Treatment Plant Status The tank detail page in Figure 9 shows live data for tank level (SCADA), water quality date from a water quality sample data base, asset data from the hydraulic model (street address, capacity,etc), live data on valve status from SCADA and forecast data from the hydraulic model on tank level. Figure 9: Tank Detail Page
BENEFITS REALISATION These decision support tools will help SA Water and Allwater staff to: efficiently manage and minimise the impact of water supply disruptions manage demands, production planning and optimise source water usage give more certainty to deferral of capital expenditure through better utilisation of existing assets provide improved knowledge and transparency within the overall water supply system The following measures and indicators were identified by the project team to demonstrate the specific benefits that can be attributable to the decision support tools: number of repair / shutdown issues escalated to significant or major events number of water quality (chlorine) events overall risk of major water supply shutdown events occurring overall water supply security risk risk of not fully realising asset capacity and therefore capital deferral benefits that can be achieved supporting continuous improvement for regulator reporting measures: o customer complaints o timeliness of response to customer complaints o water infrastructure reliability o timeliness of water service restoration In summary, the decision support tools offer significant opportunity to improve on-going management of the water supply and associated risks in the Metropolitan Adelaide Water Supply System CONCLUSION These tools are unique in that they enable realtime operational analytics what is happening now across the network and how should we respond to it but also predictive analytics what will happen in the future. Combined, these tools give SA Water access to a wealth of information not previously available and also easy access to existing information. These tools have the potential to create significant operational efficiencies, in turn delivering customer service improvements and minimising operational and capital costs. Other benefits to be realised include reduction in customers impacted by events, improved water quality event detection, improved reliability and transparency in decision making, and real time modelling and response to emergency incidents. Added to this, the depth and breadth of data generated by these tools will make regulatory compliance a much easier task for both regulators and utilities. ACKNOWLEDGMENTS The authors would like to thank SA Water for their support and guidance throughout the project and all the SA Water and Allwater staff involved in the testing and final delivery of the tools to the Operations Control Centre. There are benefits to be gained from reduction in risk and improved performance in managing shutdowns and responding to water quality events. In some areas the benefit is significant, such as reducing the likelihood of events escalating to major events, or the management of chlorine related water quality events. The escalation of events to significant or major events has a significant impact on overall customer satisfaction figures; therefore, the most significant benefit identified has been the likely improved customer satisfaction when using the decision support tools, compared to an operating environment without these tools in place. While improved operational knowledge is likely to have longer term asset management and operational cost benefits, the quantity of these benefits can only be established over time and cannot be reliably estimated at this stage.