Optimising Fuzzy Neural Network Architecture for Dissolved Oxygen Prediction and Risk Analysis

Abstract

Abstract:A fuzzy neural network method is proposed to predict minimum daily dissolved oxygen concentration in the Bow River, in Calgary, Canada. Owing to the highly complex and uncertain physical system, a data-driven and fuzzy number based approach is preferred over traditional approaches. The inputs to the model are abiotic factors, namely water temperature and flow rate. An approach to select the optimum architecture of the neural network is proposed. The total uncertainty of the system is captured in the fuzzy numbers weights and biases of the neural network. Model predictions are compared to the traditional, non-fuzzy approach, which shows that the proposed method captures more low DO events. Model output is then used to quantify the risk of low DO for different conditions.

Data Driven Models as A Powerful Tool to Simulate Emerging Bioprocesses: An Artificial Neural Network Model to Describe Methanotrophic Microbial Activity

“…DDM have the potential to simulate any biological or methanotrophic culture performance whilst considering all the needed factors that may affect the culture performance. Interestingly, these factors can be considered and simulated in a DDM, even though, the impact of these conditions are not fully understood or can be mathematically described (Khan and Valeo, 2016a, 2017a, 2017b. In other words, DDMs allow to model a system based on the available input data (e.g., the KBM coefficients) and the required output data (e.g., methanotrophic biomass).…”

“…Similarly, identifying the amount of data used for training, validation, and testing is a key factor as using a small amount of data for training can result in poor model performance and using more data will result in less data set for validating and testing (which may be insufficient for model inference). Typically, both NH and amount of data used for data division are selected based on a trial and error method, however a coupled method to identify the optimum NH and data division has been proposed (Khan and Valeo, 2017b). In this research, the previously suggested method has been followed with some minor adjustments to fit our model and is described below.…”

An Overview of Opportunities for Machine Learning Methods in Underground Rock Engineering Design

“…ANN model uncertainties come from the choice of ANN architecture (i.e., number of hidden layers, number of neurons, choice of activation function, type of training algorithm and data partitioning), as well as the performance metric chosen. Due to the data-driven nature of these models, propagating these uncertainties is easier [78,79]. One method of quantifying this uncertainty is by using fuzzy numbers to quantify the total uncertainty in the weights, biases and output of the ANN [78,80,81].…”

“…Due to the data-driven nature of these models, propagating these uncertainties is easier [78,79]. One method of quantifying this uncertainty is by using fuzzy numbers to quantify the total uncertainty in the weights, biases and output of the ANN [78,80,81]. This technique is useful for dealing with limited or imprecise datasets and can be used to conduct risk analysis [82].…”

“…As shown in Table 1, several authors have found success applying ANNs to model complex rock mass behaviour over other algorithms, which is why they are emphasized herein. To date, the most common ANNs researchers are using are MLPs with one or two hidden layers [78]. The main advantage of ANNs is that geometrical and physical constraints that govern rock mass constitutive behaviour and cause problems in numerical modelling approaches are not as problematic.…”