Papers by Niels Kjølstad Poulsen
Automated monitoring and detection of oestrus in dairy cows is attractive for reasons of economy ... more Automated monitoring and detection of oestrus in dairy cows is attractive for reasons of economy in dairy farming. While high performance detection has been shown possible using high-priced progesterone measurements, detection results were less reliable when only low-cost sensor data were available. Aiming at improving detection scheme reliability with the use of low-cost sensor data, this study combines information from step count and leg tilt sensors. Introducing a lying balance for the individual animal, a novel change detection scheme is derived from observed distributions of the step count data and the lying balance. Detection and hypothesis testing are based on generalised likelihood ratio optimisation combined with time-wise joint probability windowing based on the duration of oestrus and oestrus intervals. It is shown to be essential that cow-specific parameters and test statistics are derived on-line from data to cope with behaviours of individuals. Performance is validated on 18 sequences of data where definite proof of prior oestrus was available in form of subsequent pregnancy. These data were extracted from data sequences from 44 dairy cows over an 8 months period. The results show sensitivity 88.9% and error rate 5.9.%, which is very satisfactory when only cheap sensor data are used.
In many biomedical applications, process noise is known to be neither white nor normally distribu... more In many biomedical applications, process noise is known to be neither white nor normally distributed. When identifying models in these cases, it may be more effective to minimize a different penalty function than the standard sum of squared errors (as in a least-squares identification method). This study investigates model identification based on two different penalty functions: the 1-norm of the prediction errors and a Huber-type penalty function. In certain realistic situations, model identification based on these latter two penalty functions is shown to result in more accurate estimates of parameters than the standard least-squares solution, and also more accurate model predictions for independent test data. The effects of model identification based on these three methods are investigated in this paper. In particular, two measures of model accuracy are quantified: 1) the accuracy with which the model parameters are estimated, and 2) the accuracy of model predictions for test data...
Proceedings of the 37th IEEE Conference on Decision and Control (Cat. No.98CH36171)
In many practical systems there is a delay in some of the sensor devices, for instance vision mea... more In many practical systems there is a delay in some of the sensor devices, for instance vision measurements that may have a long processing time. How to fuse these measurements in a Kalman lter is not a trivial problem if the computational delay is critical. Depending on how much time there is at hand, the designer has to make trade o s between optimality and computational burden of the lter. In this paper various methods in the literature along with a new method proposed by the authors will be presented and compared. The new method is based on \extrapolating" the measurement to present time using past and present estimates of the Kalman lter and calculating an optimal gain for this extrapolated measurement.
Electric Power Systems Research
IFAC-PapersOnLine
We use state-based stochastic greybox modeling-combining physics and statistics-to model the slug... more We use state-based stochastic greybox modeling-combining physics and statistics-to model the slugging phenomenon. We extend the model of DiMeglio et al. (2010) to include random components and variable flow coefficients, providing 30 seconds prediction intervals. Altogether six models, each comprising no more than ten equations, are fitted to offshore riser training data and then cross-validated on new data sets. We use advanced statistical methods to 1) obtain optimal parameters of a given model fitted to measurements, 2) give model predictions with uncertainty intervals, and 3) quantitatively measure the relative goodness of the extended models. These features of our reductive method are general and can be applied to any data sets. For the slugging data, simpler models are preferable over the more complex ones (although the differences are minute for practical purposes in oil and gas industry) and a high statistical significance obtained on the training data does not imply improved long term prediction on independent data. Better physical (mechanistic) models to capture slugging oscillations are needed, ultimately to develop effective control strategies.
Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering
Gas bearing systems have extremely small damping properties. Feedback control is thus employed to... more Gas bearing systems have extremely small damping properties. Feedback control is thus employed to increase the damping of gas bearings. Such a feedback loop correlates the input with the measurement noise which in turn makes the assumptions for direct identification invalid. The originality of this article lies in the investigation of the impact of using different identification methods to identify a rotor-bearing systems’ dynamic model when a feedback loop is active. Two different identification methods are employed. The first method is open loop Prediction Error Method, while the other method is the modified Hansen scheme. Identification based on the modified Hansen scheme is conducted by identifying the Youla deviation system using subspace identification. Identification of the Youla deviation system is based on the Youla–Jabr–Bongiorno–Kucera parametrisation of plant and controller. By using the modified Hansen scheme, identification based on standard subspace identification met...
IFAC-PapersOnLine
The purpose of this study is to develop an adaptive unscented Kalman filter (UKF) by tuning the m... more The purpose of this study is to develop an adaptive unscented Kalman filter (UKF) by tuning the measurement noise covariance. We use the maximum likelihood estimation (MLE) and the covariance matching (CM) method to estimate the noise covariance. The multi-step prediction errors generated by the UKF are used for covariance estimation by MLE and CM. Then we apply the two covariance estimation methods on an example application. In the example, we identify the covariance of the measurement noise for a continuous glucose monitoring (CGM) sensor. The sensor measures the subcutaneous glucose concentration for a type 1 diabetes patient. The root-mean square (RMS) error and the computation time are used to compare the performance of the two covariance estimation methods. The results indicate that as the prediction horizon expands, the RMS error for the MLE declines, while the error remains relatively large for the CM method. For larger prediction horizons, the MLE provides an estimate of the noise covariance that is less biased than the estimate by the CM method. The CM method is computationally less expensive though.
IFAC-PapersOnLine
Users may download and print one copy of any publication from the public portal for the purpose... more Users may download and print one copy of any publication from the public portal for the purpose of private study or research. You may not further distribute the material or use it for any profit-making activity or commercial gain You may freely distribute the URL identifying the publication in the public portal If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
IFAC-PapersOnLine
Demand response (DR) will be an inevitable part of the future power system operation to compensat... more Demand response (DR) will be an inevitable part of the future power system operation to compensate for stochastic variations of the ever-increasing renewable generation. A solution to achieve DR is to broadcast dynamic prices to customers at the edge of the grid. However, appropriate models are needed to estimate the potential flexibility of different types of consumers for day-ahead and real-time ancillary services provision, while accounting for the rebound effect (RE). In this study, two RE models are presented and compared to investigate the behaviour of flexible electrical consumers and quantify the aggregate flexibility provided. The stochastic nature of consumers' price response is also considered in this study using chanceconstrained (CC) programming.
IFAC Journal of Systems and Control
Journal of Process Control
International Journal of Control, Automation and Systems
Recently research into active gas bearings has had an increase in popularity. There are several f... more Recently research into active gas bearings has had an increase in popularity. There are several factors that can make the use of gas bearings favourable. Firstly gas bearings have extremely low friction due to the usage of gas as the lubricant which reduce the needed maintenance. Secondly gas bearings is a clean technology which makes it possible to use for food processing, air condition and applications with similar requirements. Active gas bearings are therefore useful for applications where downtime is expensive and dirty lubricants such as oil are inapplicable. In order to keep as low downtime as possible it is important to be able to determine when a fault occurs. Fault diagnosis of active gas bearings is able to minimize the necessary downtime by making certain the system is only taken offline when a fault has occurred. Usually industry demands the removal of any sensor redundancy in systems. This makes it impossible to isolate faults using passive fault diagnosis. Active fault diagnosis methods have been shown able to isolate faults when there is no sensor redundancy. This makes active fault diagnosis methods relevant for industrial systems. It is in this paper shown possible to apply active fault diagnosis to diagnose parametric faults on a controllable gas bearing. The fault diagnosis is based on a statistical detector which is able to quantify the quality of the diagnosis scheme.
IFAC-PapersOnLine
The aim of this study is to develop an algorithm for detection of unannounced meals and an insuli... more The aim of this study is to develop an algorithm for detection of unannounced meals and an insulin bolus calculator (BC) to work in combination with the meal detector. The input of the meal detector are the continuous glucose monitoring (CGM) data and the insulin infusion rate. During daytime, the automated meal detector and the BC control the blood glucose concentration. During nighttime, a model predictive control (MPC) algorithm regulates the basal insulin rate. The meal detector detects the occurrence of a meal, estimates the amount of carbohydrate (CHO) in the meal, and estimates the meal onset time. The BC computes a bolus dose to cover the detected meal. We test the meal detector and the BC on nine virtual type 1 diabetes (T1D) patients. The meal detection algorithm, applied on the virtual patients, has a median detection delay of 40 min, detection sensitivity of 80% and a median meal onset estimation bias of 15 min. The algorithm does not have false positive.
Wind Energy
This paper proposes a method for real-time estimation of the possible power of an offshore wind p... more This paper proposes a method for real-time estimation of the possible power of an offshore wind power plant when it is down-regulated. The main purpose of the method is to provide an industrially applicable estimate of the possible (or reserve) power. The method also yields a real-time power curve, which can be used for operation monitoring and wind farm control. Currently, there is no verified approach regarding estimation of possible power at wind farm scale. The key challenge in possible power estimation at wind farm level is to correct the reduction in wake losses, which occurs due to the down-regulation. Therefore, firstly, the 1-second wind speeds at the upstream turbines are estimated, since they are not affected by the reduced wake. Then they are introduced into the wake model, adjusted for the same time resolution, to correct the wake losses. To mitigate the uncertainties due to dynamic changes within the large offshore wind farms, the algorithm is updated at every turbine downstream, considering the local axial and lateral turbulence effects. The PossPOW algorithm uses only 1-Hz turbine data as inputs and provides possible power output. The algorithm is trained and validated in Thanet and Horns Rev-I offshore wind farms under nominal operation, where the turbines are following the optimum power curve. The results indicate that the PossPOW algorithm performs well; in the Horns Rev-I wind farm, the strict power system requirements are met more than 70% of the time over the 24-hour data set on which the algorithm was evaluated. KEYWORDS available power, possible power, real-time power curve, real-time wake modelling 1 INTRODUCTION The share of offshore wind power is continuously increasing, especially in the Northern European grids. Together with other renewables, the accelerated implementation of offshore wind power implies many technical challenges. Particularly that the electricity system needs to adjust to the decentralized and highly variable production. To assure safety in the operation of the power system, offshore wind farms are designed as wind power plants (WPPs) required to contribute to the stability of the grid by offering grid services (also called ancillary services). As part of those services, offshore WPPs provide operating reserve capacity to the electricity network, which is activated by down-regulating the wind farm from its maximum possible power. 1-3 The estimation of the available power, or eventually the reserve capacity, is essential as the balancing responsible parties are compensated for this service in terms of the level of reserves, regulated by the national grid code. The reserve power can also be traded in the balancing market, depending on the regional market schemes. What is seen from the existing European regulations 4 is that adequate and standardized regulations or technical requirements to help in understanding the possible power or the amount of reserves for their system reliability is lacking. This research is critical not only for the power system stability but also for the business case of wind energy.
Biomedical Signal Processing and Control
Computing & Control Engineering Journal
IFAC-PapersOnLine
This paper addresses model identification of continuous-discrete nonlinear models for people with... more This paper addresses model identification of continuous-discrete nonlinear models for people with type 1 diabetes using sampled data from a continuous glucose monitor (CGM). We compare five identification techniques: least squares, weighted least squares, Huber regression, maximum likelihood with extended Kalman filter and maximum likelihood with unscented Kalman filter. We perform the identification on a 24-hour simulation of a stochastic differential equation (SDE) version of the Medtronic Virtual Patient (MVP) model including process and output noise. We compare the fits with the actual CGM signal, as well as the short-and long-term predictions for each identified model. The numerical results show that the maximum likelihood-based identification techniques offer the best performance in terms of fitting and prediction. Moreover, they have other advantages compared to ODE-based modeling, such as parameter tracking, population modeling and handling of outliers.
Journal of Process Control
Journal of Process Control
Uploads
Papers by Niels Kjølstad Poulsen