loading is 5-day BOD, which takes more than 5 days to obtain. It is, therefore, almost useless for automatic control except on a seasonal basis for actions such as bringing basins online or taking them offline;
• Water resource recovery facilities seldom, if ever, operate with constant or constrained flows or loadings. Facilities experience significant variations on a diurnal and seasonal basis and minute-to-minute variations in influent composition. There are even significant temporal differences between inputs on weekdays compared to weekends. With such large influent variations, it is typically not possible to control all process outputs of interest to within desirable limits;
• Real-time process models are not as well defined for wastewater processes as they are for many industrial processes. The most accurate deterministic models available have inputs currently immeasurable in real time;
• Many WRRFs are designed for future conditions and have control elements (valves, pumps, etc.) that are oversized for current loadings. This type of overdesign often leads directly to poor control; and
• Traditionally, most public utilities have bid projects and awarded contracts to the low bidder. This practice led to many “or equal” substitutions that are really not equal. The result has been a history of automation failures.
This practice has led to risk-adverse behavior, and it has fostered an unwillingness to innovate. On the positive side, wastewater treatment processes have several factors that make advanced control attractive:
• Advanced wastewater processes with many interacting control handles are often difficult to control manually. They use processes with long delay times and time constants that can vary from seconds (blower control) to minutes (dissolved oxygen control) to days (sludge wasting) for the various treatment processes; and
• Automation coupled with advanced control strategies has the potential to both significantly improve process performance and decrease energy use.
A review of the literature revealed a number of papers on model-based control. Most were based on theoretical development of controllers or were used in simulation only. Only a few papers appeared to present full-scale applications. However, those few applications showed promising results. And, with so few of the nation’s wastewater facilities currently automated, there is a tremendous opportunity for improvement through the use of model-based control.
4.9 Artificial Neural Networks
During the past two decades, a series of sophisticated models such as ASM1, ASM2 (Guyer et al., 1994), ASM2D (Henze et al., 1999), and ASM3 (Guyer et al., 1999) were introduced that simulate BOD and chemical oxygen demand removal, nitrification, denitrification, and biological phosphate removal. These newer models are highly accurate and useful for design, research, and other offline activities. However, they are seldom used for online process control, mainly because several of their inputs cannot be measured in real time. Models used for real-time control must use inputs that are practical to measure online and in real time, relatively accurate, and easy to calibrate. Artificial neural network (ANN) models meet these requirements. Advantages and disadvantages of ANN models are listed in Table 7.2 (Hill, 2010).
4.9.1 Artificial Neural Network Models
Artificial neural network models are nonlinear statistical data modeling tools loosely based on biological neural networks. They can model real world systems by tuning a set of parameters. These parameters, known as weights, describe a model that maps from a set of given inputs to an associated set of outputs. The process of tuning the weights to the best values is called training. In simple terms, an ANN is a data-driven empirical model.
Artificial neural networks come in several different structures that are each most suitable for specific types of tasks. Table 7.3 lists common types of ANNs and their applications.
Figure 7.6 shows the structure and components of a back propagation ANN model. Inputs are shown on the left of the figure and outputs are shown on the far right. The hidden layer is a sequence of nodes. The arrows between the inputs and the nodes each represent a model weight that is multiplied by the input value.
TABLE 7.2 Advantages and disadvantages of ANN models (Hill, 2010).
TABLE 7.3 Types of ANNs and their application areas.
Each node in the hidden layer and the output layer has a structure and functions as shown in Figure 7.7. The first function is a summation of the inputs multiplied by their respective weights. The activation function can take any form, but is most commonly monotonic. Typical activation functions include linear, hyperbolic tangent, sigmoid, step, and exponential.
FIGURE 7.6 Components of a back propagation ANN model.
FIGURE 7.7 Functions of a neural network node.
The weights of an ANN model must be calibrated. This process is called training and involves running an input and output data set through the model and incrementally improving the estimate of the weights. Training is computationally intensive, but is typically one of the easier steps in model identification. There are many software packages available with this capability.
4.9.2 Artificial Neural Network Models Used for Process Modeling
Artificial neural network models, both with and without feedback, can be used for process simulation. Figure 7.8 presents a schematic of the structure of the model without feedback. The input, u(t–n), is a vector of parameters at one or more past times. These parameters might include influent conditions such as flowrate and ammonia concentration and other measured values such as dissolved oxygen concentration and mixed liquor suspended solids concentration. The output, ŷ(t), is a vector of parameters at time, t, and might include ammonia and nitrate concentrations. Thus, the ANN model takes the inputs from past times and calculates the outputs for one time step ahead.
FIGURE 7.8 Artificial neural network model without feedback.
FIGURE 7.9 Artificial neural network model with feedback.
Figure 7.9 presents a schematic of the ANN model with feedback. In addition to the input vector, u, the model uses the value of the measured outputs at the past time period, y(t–1), as an input. The overall effect is that this type of model only needs to calculate the change in output parameters because it is constantly being updated with the most recent measured output information.
Determining
