Introduction This paper discusses theimplementation of artificial intelligence technology (technology with thecapability of solving problems which would normally require human intelligence)into the selection and manipulation processes of flow and water quality modelsin coastal environments. Currently, various numerical models (typically, modellinga situation by observing its behaviour over time and manipulating the datamathematically) are used to solve flow and water quality problems. However, theselection of a model is a very specialised and difficult task, requiringsignificant knowledge of the applications and limitations of models, modelfeatures must match with the particular situation. In addition, the manipulationof certain models requires an experienced user.
Therefore, due to a lack ofknowledge of factors such as water depth, water velocity, grid spacing etc., inexperiencedusers will have difficulty with these tasks. Artificial Intelligence has theability to store knowledge to aid all users with the selection and manipulationof mathematical models. Need to integrate with Artificial Intelligence The manipulation of the models isnecessary to account for changes in initial parameters (as results may beskewed if these changes aren’t made). Most model users lack the knowledge tounderstand, model and evaluate the data and subsequently models can be misused,underutilized and can even lead to incorrect conclusions. Within modelmanipulation, the aim is to achieve a good standard of simulation. Model usersare able to modify one or two parameters of a model to suit a certain situationbut any more and there can be confusion.
AI techniques are able to replicatethis and can cope with further manipulations, maximising the effectiveness ofcurrent models. The latest generation of models can integrate AI technologywith computational methods to solve and analyse issues with fluid flows to forma single system to aid non-experienced users. Integration with Artificial Intelligence Theintegration of AI technology will provide an effective decision-making tool andwill quicken the water quality planning and control process, by providing faster,more effective and more organised (better storage, retrieval and manipulationof data) models.
4 AI algorithms have been explored, KBSs, GAs, ANNs and FISs,these are explained below: 1 – Knowledgebased systems (KBSs)Interactive,user friendly computer programs that replicate the decision making andevaluation processes of experts in solving a specific problem, throughdelivering advice, answering questions and justifying conclusions. Most of theKBSs knowledge is derived from expert knowledge which is stored in a knowledgebase from which inferences are made based upon responses to certain questions. Anexample of a KBS in water flow and quality modelling are a decision-supportsystem for river basin planning and management.
Current models only aid inmodel selection and not in manipulation, selection alone requires a largeeffort to program, with more complex issues the integration of KBSs becomesvery complicated.2 –Genetic algorithms (GAs)Theseare ‘evolutionary optimization algorithms’ that utilise models of theevolutionary process to develop a problem solving system. They use biologicallyinspired search techniques like natural selection, reproduction, crossover andmutation to determine the optimum response. A knowledge base (including priorknowledge and knowledge of all previous solutions deduced by the system) iscoupled with an artificial survival of the fittest logic system to choose thebest solution.
An example of an application is the use of GA to optimizewastewater treatment in a water quality management model. These algorithms arehighly effective at solving complex, multi-parameter problems. 3 – Artificial neuralnetworks (ANNs)Algorithms basedon the current understanding of the brain and its nervous systems, the innerworkings of an ANN are outlined in Figure1.
Some issues with ANNs are that they utilise all environmental parametersrather than the most important parameters in water quality. Additionally,little is done to extract information from the network’s learning system, so itis often not specified how conclusions are reached. This technique requires furtherdevelopment in relation to water quality problems. An example of an applicationis where an ANN was used to optimize watershed management for a reasonablebalance between water quality demand and farming industry restrictions.
4 – FuzzyInference systems (FISs)Thisalgorithm specialises in modelling complex and vague systems when theconstraints and objectives are unclear. Data entering the system is mapped tovalues that the fuzzy knowledge base can understand, the logic base thenapplies an ‘if – then’ concept basing its decision from its knowledge base, todetermine the outcome. Essentially, in a fuzzy system a computer may notprovide a yes or no question with a yes or no answer and will provide ajustification of the response, similar to what humans do at times. An exampleof a FIS is in the design of a algae bloom predictor based on the dailyfluctuations of water quality parameters (e.g.
pH, temperature, dissolvedoxygen etc.). Conclusion In conclusion, current water qualitymodels have a variety of constraints and are difficult to use in terms of modelselection and manipulation, especially for inexperienced users. AI technologiesprovide the ability to apply expert knowledge to situations without thepresence of an expert model-user. KBSs, GAs, ANNs and FISs are possible AItechnologies that can be used in water quality modelling and to decide whether numericalmodels represent actual phenomena. In further development it is projected thattwo or more of these systems can be combined to produce an even better waterquality modelling system.
Moreover, better user interfaces and more efficientAI technologies are in the works.