Water advisory demand evaluation and resource toolkit
Cities are living organisms, 24h / 7day, with demands on resources and outputs. Water is a key resource whose management has not kept pace with modern urban life. Demand for clean water and loads on waste water no longer fit diurnal patterns; and they are impacted by events that are outside the normal range of parameters that are taken account of in water management. This feasibility study will determine how the application of computational intelligence can be used to analyse a mix of data inputs to produce credible predictions for clean water demand and foul water outputs in urban areas. The data inputs will be social-media and gas and electricity usage, combined with meteorological and traffic movement data. These will deliver predictions of population density and activity over a subsequent 8 hours period, thus providing inputs to the water supply services on the future demand of fresh water supplies, and the subsequent load on waste water and sewerage systems. The innovation of this concept is the aggregation of social-media data with transport related data to deliver a toolkit that predicts population density in an urban area over the next 8 hours. The toolkit will output the predictions in an open-source manner to support interoperability; thus enabling the development of new applications. For the sake of feasibility study the obtained data sets are localised to Leicester city in United Kingdom. The created online database contains mix of historic and real-time data. Data sources which are monitored and collected in real-time are localised Twitter feeds, current gas and electricity usage on regional level, traffic information from in-situ sensors and from traffic monitoring institutions, weather forecast and rainfall data. To ease the work with such large dataset a graphical user interface was developed in Matlab software and employed capabilities its specialised toolboxes. The online database is based on the Microsoft Azure solution. The computational intelligence model currently developed consist of various topologies of artificial neural networks and support vector machine regression. Note that the ﬁnal model will comprise at least two models with weighted outputs as initial studies suggested that one model may not capture all the possible trends that characterises the training data for artificial neural network. The created toolkit includes a sensitivity test unit to evaluate the importance or contribution of each of the input variable on the prediction accuracy of the model, and also as a means of comparing our approach with traditional methods of population and water prediction. The toolkit aims to provide predictions for different time intervals, e.g. hourly, daily, monthly and yearly. Embedded within the tool are variants of differential evolutionary and swarm intelligence optimisation algorithms for optimising the meta-parameters of the computational intelligence models and the weights of the combined model. To test the functionality of the developed tool along with appropriateness of the proposed approach for the water demand prediction, data obtained from the SmartSpaces website (http://smartspaces.dmu.ac.uk) were utilised. This website shows the energy performance of a selection of public buildings in Leicester such as De Montfort University campus buildings, Leicester City Council buildings, schools, libraries, leisure centres and others buildings in Leicester. The SmartSpaces website monitors at 30 min intervals temperature, usage of electricity, gas and water within the buildings on the list. While the number of monitored buildings on the SmartSpaces website is limited, it provided a convenient access to the data and thereby enabled development of initial models. For testing the functionality of the toolkit using historic data from the SmartSpace project, the inputs of the artificial neural network and support vector machine models include electricity, gas, temperature and two recent past water demand. The output is the predicted current water demand. The outcomes from the initial study seem promising as the water usage was predicted with an average mean square error of 0.119 in terms of cubic meters.
Research Group : DIGITS
Research Institute : Institute of Artificial Intelligence (IAI)
Peer Reviewed : Yes