Evaluation of the impact of technological interventions on the quality of the water in the canals, with the application of a neural network model
The chemical-physical data acquired in various tide and seasonal conditions based on the methods of sampling defined in the protocol by Cnr Ismar of Venice (the document “Progetto Icaro – phases 1 and 2 – results and methodologies”) have been used to supply a neural network model (Renp).
A neural network is a group of simple but highly interconnected elaboration units (neurons), that interact among themselves and with objects from the outside world by exchanging signals in a way that is very similar to that of biological neural structures.
Basically the neural networks are adaptive systems that “learn” to execute a certain task correctly by following examples. The information contained in the examples is used to adjust the collection of parameters inside the network that can later influence how it works once it becomes operative. The main fields of application are the solutions of problems in classification, clustering, optimization and prediction. An artificial neural network is a simplified model of the human central nervous system. It is composed of a certain number of base units, known as neurons, that can communicate among themselves thanks to connections known as synapses. These can be organized in various ways, depending on the type of network one wants to create. Starting with examples provided as input, the network successively elaborates its own inner representation of the data, which then leads to the desired output.
In synthesis, neural networks make it possible to study and reproduce very complex functions. Special training techniques make it possible to approximate and provide a precise replica of the development of certain phenomena.
The neural model used in the study was implemented and calibrated on a collection of hydrodynamic and chemical data acquired in the system “Borgoloco Pompeo Molmenti”, during the 2002 and 2003 Icaro projects and was the perfect tool for the analysis and assessment of the conditions of a significant series of chemical, chemical-physical and hydrodynamic parameters which would make it possible to complete an environmental classification of the canals in Venice.
The database of the reference data is constituted by a collection of up-to-date current measuring data (speed of the current and water level) provided by an electromagnetic current meter installed in the Rio del Piombo, by a series of specific speed and water level measurements acquired by a predetermined group of measuring stations, and by series of measurements of the various chemical types of nutrients, previously acquired by taking samples of the water from the Rio del Piombo in various water level and seasonal conditions.
The elaborations produced demonstrate the excellent functionality of the neural model implemented in the prediction of variables determined by non-linear environmental processes.
The mistakes made in predicting chemical variables using the definitive calculation algorithm were particularly insignificant in terms of the median value and variance, with a normal type of frequency distribution. The greatest errors were committed in the simulation of the nutrients in the Canal of San Zulian, because of the exceptional environmental complexity of that system and the imperfect set of calibration data that was used. In the Rio del Piombo and in the canals of the system of the islands of San Vio – Santo Spirito the simulation was particularly satisfactory. The improvements in the model’s algorithm made it possible to make excellent simulations even for the relatively specific values of the series under analysis.
The neural model may be applied not only for the prediction of chemical and hydrological variables, but also for the simulation of other types of parameters, such as for example variables in the amplitude of waves. The finished simulation also demonstrated that it predictions could be made with particularly irrelevant errors.