Wenkai Hu

A new method to detect and quantify correlated alarms with occurrence delays

Abstract This paper proposes a new method to detect correlated alarms and quantify the correlation level to improve the management of industrial alarm systems. The method is mainly composed of three parts. First, a so-called occurrence delay is defined as the main cause leading to erroneous conclusions from existing methods to detect correlated alarms. In order to tolerate the presence of occurrence delays, a mechanism is presented to generate continuous-valued pseudo alarm signals. Second, a novel approach is given to estimate the correlation delay between alarm signals, so that the correlation delay can be separated from occurrence delays to obtain real occurrence delays (ROD). Third, a statistical test based on the ROD is proposed to determine whether two alarm signals are correlated or not, and the Pearsons correlation coefficient is applied to quantify the correlation level. Numerical examples and industrial case studies are provided to support the proposed method.

Discovering Association Rules of Mode-Dependent Alarms From Alarm and Event Logs

State-based or condition-based alarming has emerged as a prevalent method to reduce nuisance alarms and inhibit alarm floods in the alarm management of process industries. Such a strategy minimizes the number of active alarms by modifying alarm attributes or suppression status based on certain conditions. However, the configuration of state-based alarms in practice relies on process knowledge, making it time and resource intensive. In order to identify associations between alarms and states, this paper proposes a completely automated data-driven method to detect mode-dependent alarms from alarm and event (A&E) logs, where the messages of alarms and operating modes are stored. Algorithms to detect frequent patterns of operating modes and association rules of mode-dependent alarms are proposed. The effectiveness and applicability of the proposed method are demonstrated by case studies involving real industrial A&E data sets.

Multimodel Parameters Identification for Main Steam Temperature of Ultra-Supercritical Units Using an Improved Genetic Algorithm

AbstractA parameter identification method based on genetic algorithm (GA) is presented to solve the multimodel parameters identification problem for the main steam temperature of ultra-supercritical (USC) units. Linear ranking selection, nonuniform linear crossover, and Gaussian mutation are employed in the algorithm design to enhance the convergence speed and the accuracy of the identification. Besides, the uniform design method is executed to initialize the population and the sigmoid function with adaptation is applied to adjust the probabilities of crossover and mutation. Simulations carried out with the field operation data from Haimen USC units, including two processes—parameters identification and model verification. The simulation results show that the improved genetic algorithm performs well in global parameters searching and the proposed identification methodology offers good results for the multimodel parameters identification.

An automated data-driven method to detect mode-based alarms

A variety of alarm management techniques are available to improve the performance of alarm systems and avoid alarm overloading; in particular, the state-based alarming strategy has been widely used in practice to remove noninformative alarms that are caused by the switching of operating modes. However, the configuration of mode-based alarming strategies relies on proficient process knowledge, and thus is time and resource intensive. To address this problem, this paper presents a completely automated data-driven technique to detect mode-based alarms from historical Alarm & Event (A&E) logs. The major contributions are: 1) the detection of mode-based alarms is formulated as a hypothesis testing problem; 2) systematic detection methods are proposed to process A&E data and output final results as association rules. The efficacy of the proposed method is illustrated by industrial case studies involving real A&E data.

Framework for a smart data analytics platform towards process monitoring and alarm management

Abstract The fusion of information from disparate sources of data is the key step in devising strategies for a smart analytics platform. In the context of the application of analytics in the process industry, this paper provides a framework for seamless integration of information from process and alarm databases complimented with process connectivity information. The discovery of information from such diverse data sources can be subsequently used for process and performance monitoring including alarm rationalization, root cause diagnosis of process faults, hazard and operability analysis, safe and optimal process operation. The utility of the proposed framework is illustrated by several successful industrial case studies.

Improved Genetic Algorithm Based Predictive PID Control for Main Steam Temperature of Ultra-Supercritical Boiler

A main steam temperature control method of ultra-supercritical boiler which can optimize the parameters of predictive PID controller online by using an improved genetic algorithm is proposed in this paper. To solve the existing problems of the constraint solving control, the optimization idea of genetic algorithm based on generalized predictive control performance index is proposed and the PID parameters optimization models of the inside and outside circuit are established. Sorting and optimal guaranteed option, non-uniform linear crossover and Gaussian mutation operator are introduced into genetic operators, the initial population uniform design method is adopted and the probability of crossover and mutation is adjusted adaptively by Sigmoid function, in order to improve the convergence accuracy and speed of genetic algorithm. A control simulation is conducted on the historical data of the continuous operation of Chaozhou Power Plant 3# ultra-supercritical boiler site, and the results show that the system which adopts the prediction PID control strategy based on improved genetic algorithm has better dynamic and static characteristics and also can better meet the requirements of full operating conditions than the system with traditional control strategy.

Study on the layered predictive control algorithm for the main steam temperature of ultra-supercritical units

Aiming at the main steam temperature control problem of ultra-supercritical units under variable loads conditions, a predictive PID control algorithm with layered structure is proposed. The cascade PID control is applied in the lower circuit which can realize the application in DCS system through configuration language programming. The generalized predictive optimization is applied in the upper circuit which is used to conduct PID parameter setting instead of control engineers. By introducing least square method with forgetting factor recursive, the model parameter identification in rolling window is realized and the PID parameters tuning optimization model with predictive index is established. The simulation results show that the proposed algorithm can adapt to the model change process of the controlled object of main steam temperature and is qualified with strong stability and robustness.