Jaakko Talonen

Self-organizing map in process visualization

Our research group is studying data-analysis based techniques in decision support and visualization. We have had co-operation with a Finnish nuclear power plant Olkiluoto within a long industrial research project. We have developed many decision support schemes and visualizations based on self-organizing map (SOM) method combined with other methodologies. In this paper, we discuss about SOM method in the process visualization of dynamic systems. With a case example produced with the Olkiluoto plant data we show the information value of this method. Some comparisons to other methodologies are made and the assessment of the information value and the definition of the assessment criteria are discussed. The measurement of the information value is a challenging task.

Combining neural methods and knowledge-based methods in accident management

Accident management became a popular research issue in the early 1990s. Computerized decision support was studied from many points of view. Early fault detection and information visualization are important key issues in accident management also today. In this paper we make a brief review on this research history mostly from the last two decades including the severe accident management. The authors studies are reflected to the state of the art. The self-organizing map method is combined with other more or less traditional methods. Neural methods used together with knowledge-based methods constitute a methodological base for the presented decision support prototypes. Two application examples with modern decision support visualizations are introduced more in detail. A case example of detecting a pressure drift on the boiling water reactor by multivariate methods including innovative visualizations is studied in detail. Promising results in early fault detection are achieved. The operators are provided by added information value to be able to detect anomalies in an early stage already. We provide the plant staff with a methodological tool set, which can be combined in various ways depending on the special needs in each case.

New visualization techniques and their assessment

Our research group has been studying data-analysis based techniques in decision support and visualization. We had a long industrial research project in co-operation with a Finnish nuclear power plant Olkiluoto. We developed many decision support schemes based on Self-Organizing Map (SOM) method combined with other methodologies. Also several visualizations based on various data-analysis methods were developed. Data from the Olkiluoto plant and training simulator was used in the analysis. In this paper some of these visualizations are presented, analyzed, and assessed with a psychological framework. Measuring the information value of the visualizations is a real challenge. The developed visualizations and visualization techniques are also compared with some existing visualizations and techniques in current plants and research laboratories. The visualizations and the visualization techniques are developed further, and completely new visualizations and techniques are developed. We point out what additional value the new visualization techniques can produce. A detailed test case of using Self-Organizing Map (SOM) method with Olkiluoto plant data is presented. With this practical example the information value of this method is shown, and it is also pointed out how it can be assessed, and what are the most reliable criteria in this assessment.

Data fusion of pre-election gallups and polls for improved support estimates

In this paper, gallup results and a questionnaire in the context of a voting advice application related to the Finnish presidential election are combined. The main emphasis is on preprocessing phases where raw data is reformed to temporal data sets. We also pay attention to find optimized parameters for a merged recursive model. Aggregated data from a questionnaire was stored frequently and modified by a differential equation. The method presented in this paper allows us to visualize more accurately the daily support of each candidate before the election. The results can be used for further research such as forecasting the results and the success of presidential campaigns.

Modelling hypothetical wage equation by neural networks

In this paper, a hypothetical wage equation is modelled using quarterly data from United Kingdom. Wage and price data have a great importance for the overall features of large-scale macro models and for example for the different policy actions. The modelled feature in this paper is the real wage, the differential of a nominal wage and a price index. In the variable selection phase, the stationary properties of the data were investigated by augmented Dickey-Fuller tests (ADF). The main idea in this paper is to present a neural network model, which has a better fit than conventional MLR model.

Generated control limits as a basis of operator-friendly process monitoring

Early and accurate fault detection has long been a major challenge in safety of industrial plants. Fault detection and diagnosis can minimize downtime, increase the safety and reduce manufacturing costs. Industrial plants are becoming more instrumented, resulting a large amount of high-dimensional data. In this paper, an approach based on data generated control limits is suggested to detect faults or pre-stage of abnormal event in the process. A feature called alarm balance was developed to clarify monitoring at the power plant. Regardless, the final decision of the process state belong to operators. Real and simulator data is available for this research from Olkiluoto nuclear power reactors in Finland.