When should we introduce the engineered bacteria and activate the light-on killing system are questions keeping bothering us. Also, traditional water detection method has the defects such as long cycle time and inaccuracy, making it hard to maximize the function of our engineered bacteria. Therefore, software compatible with the iTube is needed to solve these problems. With Python and Matlab, we build a software integrating data acquisition and modeling to control the iTube and the behavior of the bacteria.
Corrosion is a process caused by multiple factors. To better predict the corrosion rate of a pipeline, temperature, pH, electrical conductivity and alkalinity are chosen as the input of the neural network based on the analysis of their importance to the corrosion process, while the corrosion rate is the output. All the sample data ranging from 2016 to 2018 are collected from a petrochemical company and a steel company.
Neural Network Configuration
The water quality related data is actually a time series sequence, and NARX neural network is applied in this study considering it is a suitable predictor for dynamic nonlinear system. And also given the availability of past water quality data, series-parallel architecture is ideal for the model.
After several modification and testing, 1 hidden layer and 15 neurons in the hidden layer has the best performance and minimal MSE.
Prediction Performance and Discussion
Figure 3 shows the cost function. Though the model is useful in certain circumstances, it varies from factory to factory. Before the implementation of the model, data from certain factory should be obtained and retrained for the best performance of the prediction.
Q. Gao et al., "The Research of the Water Quality Prediction Model for the Circulating Cooling Water System", Applied Mechanics and Materials, Vols. 385-386, pp. 408-411, 2013