Development of an artificial neural network-based software for prediction of power plant canal water discharge temperature

https://doi.org/10.1016/j.eswa.2005.06.009Get rights and content

Abstract

Power plant cooling water systems that interact with nearby effluents are complex non-linear, large-time-delay systems. A neural network-based software tool was developed for prediction of the canal water discharge temperature at a coal-fired power plant as a function of plant operating parameters and local weather conditions, including tide information. The plant has four units totaling an installed capacity of 1550 MW and its water thermal discharge is environmentally regulated. In the summer months, when the price of electricity is very profitable and the risk of exceeding the canal temperature limit is greater, the tradeoff between maximum generation and environmental compliance violations is financially significant. The software is a predictive tool to assist in scheduling load generation among the plant's four units without exceeding a thermal discharge limit of 95 °F. Back propagation neural network architectures were trained using plant operating data with an ‘off-set’ component. The artificial intelligence models produced reasonable trends for year-round prediction and different operational scenarios. Comparison of measured and predicted canal temperatures indicated an accuracy of less than 0.3 °F over the range between 90 and 95 °F. The software tool was developed as an Object Linking and Embedding (OLE) for Process Control (OPC) client, with real-time communication and interface with the plant Distributed Control System (DCS).

Introduction

Thermal effluents from power plants in the United States that are discharged into water pools are subject to environmental regulations to minimize its potential adverse impact on the local warm water fishery. The adjacent water temperature may become elevated above background, and the elevated water temperatures could propagate downstream along with the thermal effluent. Since, the thermal effluent is directly affected by the total power generation from the power plant, regulatory water discharge temperature limits impose a constraint on the plant available maximum electrical generation. On an unregulated power generation market, this has a direct impact on the generating utility market share and financial profits. Attempts have been made to develop theoretical models for power plant water discharge temperature, using a combination of first principle and empirical models (Kalogirou, 2000). Such models involve complex calculations using parameters that are not readily available in the station Distributed Control System (DCS). Furthermore, these models do not work well under different operational scenarios.

A study was performed to develop a software tool to predict water discharge temperature and provide advice on generating unit dispatch at a power plant subject to an environmental effluent regulation. The station consists of four power generation units that interact with the effluents from two rivers and produce a combined canal water thermal discharge, which is limited to an hourly average mandated state limit of 95 °F. The canal is 3200-ft long, so there is a large time delay from the plant discharge to the point of temperature monitoring, that makes the regulated temperature a strong function of the atmospheric conditions. The circulating water system at each unit provides cooling water for the condensers using variable speed pumps with maximum flow capacity of 921 million gallons per day. One of the units can operate on a ‘piggyback’ mode, on which, water can be recirculated from the discharge canal and used to condense steam in that unit. During the summer months, when the price of electricity is very profitable and the risk of exceeding the canal temperature limit is greater, unit loading is accomplished by the operators by using a heuristic approach based on previous experience and the results of a simplified engineering analysis.

The objective of this study was to develop software capabilities which will be used by the plant operators to optimize load generation patterns (especially during periods of high generation demand) considering unit operating data, local weather conditions, and the constraint imposed by the environmental canal thermal discharge limit. Artificial neural networks (ANN) were used to develop a predictive model of the canal water discharge temperature. The results demonstrated that this application is well suited for this area of artificial intelligence. The ANN models were built using NeuFrame v.4 from Neusciences, UK. The network models were evaluated in their ability to predict the average canal temperature as a function of plant operating parameters and atmospheric conditions, including tide information. Local weather information was provided from a weather station installed on-site. Tide information was taken from the US National Oceanic and Atmospheric Administration's Center for Operational Oceanographic Products and Services. The model demonstrated its ability to produce reasonable trends and to predict the canal temperature over the normal operating range with good accuracy. The precision of the predicted temperature, over the entire temperature range is better than ±0.5 °F. Additionally, the operative software was built to run as an Object Linking and Embedding (OLE) for Process Control (OPC) client to the station's Honeywell OPC server. The software provides pertinent on-line plant data to the server for display, and provides real-time canal temperature prediction 30-min into the future, and expert advice for unit load dispatch. Results of an evaluation over a 7-month period indicated good software functionality and robustness over time.

Section snippets

Artificial neural networks

Recent years have witnessed a rapidly growing interest in artificial neural networks and their application to various scientific and engineering domains. ANNs have proven very useful in the analysis of complex and uncertain data, a feature common in electric power system problems. A neural network is a system of interconnected processing elements, inspired by the network structure of the brain, which learns the relationship between input data vectors and the output(s). The networks are

Canal temperature neural network model

The approach used in this study employs a trained neural network that relates the discharge canal average temperature to selected plant operating parameters and local weather conditions that include the different seasons of the year. Input data, from operational months in 2001, were selected based on their impact on network model performance and pre-processed using standardization techniques. The network architecture was optimized based on the network model performance. A network model was then

Predictive software tool

A Visual Basic interface subroutine was developed around the ANN engine. The Visual Basic code handles on-line input data from the plant DCS system, any necessary calculations to query the ANN model, the real-time temperature predictions and supportive information for display. The Visual Basic code was built with capabilities to handle different ANN models that could be built in the future to account for new system configurations. An object linking and embedding (OLE) for process control (OPC)

Model validation

A log filing capability was built into the software tool. The intention of this capability was to provide data for easy evaluation of software prediction accuracy. Fig. 5, Fig. 6, Fig. 7, Fig. 8 provide a comparison between predicted and measured canal discharge temperature for the months of May, June, July and August 2002. The canal water temperature steadily increased over these months from the 60 °F level to values around the 95 °F mark. Sometimes the plant exceeds the 95 °F limit; however,

Conclusions

Artificial neural network provide a new approach to the development of thermal and fluid process models that are difficult to fully model from first principles. A study is presented where ANN were used to predict power plant thermal effluent water temperature to help optimize load generation among power plant generation units at a power plant that is subject to an environmentally regulated canal water discharge temperature limit of 95 °F. Network models were evaluated in their ability to predict

References (4)

There are more references available in the full text version of this article.

Cited by (24)

  • Artificial neural networks and physical modeling for determination of baseline consumption of CHP plants

    2014, Expert Systems with Applications
    Citation Excerpt :

    Reproducing the behavior of plant components (boilers, steam turbines, and superheaters) (Bekat, Erdogan, Inal, & Genc, 2012; De, Kaiadi, Fast, & Assadi, 2007; Du et al., 2011; Jia & Xu, 2010; Ma, Wang, & Ma, 2008; Olausson, Häggståhl, Arriagada, Dahlquist, & Assadi, 2003). Optimization of load distribution and load disconnection for stability purposes in the case of fault of the system (Hsu, Chuang, & Chen, 2011; Romero & Shan, 2005). Optimization of operating parameters for plant efficiency maximization, in some cases through combined implementation of ANN and other optimization techniques, e.g., genetic algorithm (Arslan, 2011; Hajabdollahi, Hajabdollahi, & Hajabdollahi, 2012; Rashidi, Galanis, Nazari, Basiri Parsa, & Shamekhi, 2011; Suresh, Reddy, & Kolar, 2011).

  • Back pressure prediction of the direct air cooled power generating unit using the artificial neural network model

    2011, Applied Thermal Engineering
    Citation Excerpt :

    Moreover, the ANN approach allows to simulate the physical phenomena without the explicit formulation of the possible relationships that may exist between the variables [14,15]. Recently, the ANN model was applied to predict the back pressures of a cooling tower under a broad range of operating conditions by Hosoz et al. [16], and a neural network based software tool was presented by Carlos et al. [17] to predict the canal water discharge temperature of a coal-power plant based on the operating parameters and the local weather conditions. The ANN approach has also been extensively used to investigate the energy systems by several researchers [18–24], among which, modelling on the induction motor, solar air heater and ground-coupled heat pump systems was revealed and discussed.

  • Neural model for forecasting temperature in a distribution network of cooling water supplied to systems producing petroleum products

    2011, International Journal of Refrigeration
    Citation Excerpt :

    The most frequently used model for the mapping of non-linear relationships in cooling and heat exchange processes are multi-layer perceptron (MLP) neural networks. The application of such networks is explored in the following studies: De et al. (2007), Ermis (2008), Fast et al. (2009), Islamoglu and Kurt (2004), Lu and Hogg (2000), Qi et al. (2008), Romero and Shan (2005), Rusinowski and Stanek (2007), Smrekar et al. (2009), Yigit and Ertunc (2006), Boccaletti et al. (2001), Gao et al. (2009), Hosoz et al. (2007). Apart from MLP neural networks, in studies by Navarro-Esbrí et al., (2007), Tan et al. (2009), Yao et al. (2006), the application of RBF (Radial-Basis-Function) neural networks is examined.

  • A neural network-based estimation of electric fields along high voltage insulators

    2009, Expert Systems with Applications
    Citation Excerpt :

    Many studies on power systems using ANN have been reported in literature. ANNs have proven very useful in the analysis of complex and uncertain data, a feature common in electric power system problems (Romero & Shan, 2005). An ANN model forecasting the flashover voltage of an insulator in complex ambient conditions investigated in Sima et al. (2004).

View all citing articles on Scopus
View full text