In this research work a novel two-step system for anomaly detection is presented and tested over several real datasets. In the first step the novel Exploratory Projection Pursuit, Beta Hebbian Learning algorithm, is applied over each dataset, either to reduce the dimensionality of the original dataset or to face nonlinear datasets by generating a new subspace of the original dataset with lower, or even higher, dimensionality selecting the right activation function. Finally, in the second step Principal Component Analysis anomaly detection (...) is applied to the new subspace to detect the anomalies and improve its classification capabilities. This new approach has been tested over several different real datasets, in terms of number of variables, number of samples and number of anomalies. In almost all cases, the novel approach obtained better results in terms of area under the curve with similar standard deviation values. In case of computational cost, this improvement is only remarkable when complexity of the dataset in terms of number of variables is high. (shrink)

The use of batteries became essential in our daily life in electronic devices, electric vehicles and energy storage systems in general terms. As they play a key role in many devices, their design and implementation must follow a thorough test process to check their features at different operating points. In this circumstance, the appearance of any kind of deviation from the expected operation must be detected. This research deals with real data registered during the testing phase of a lithium iron (...) phosphate—LiFePO4—battery. The process is divided into four different working points, alternating charging, discharging and resting periods. This work proposes a hybrid classifier, based on one-class techniques, whose aim is to detect anomalous situations during the battery test. The faults are created by modifying the measured cell temperature a slight ratio from their real value. A detailed analysis of each technique performance is presented. The average performance of the chosen classifier presents successful results. (shrink)

This paper aims to enhance security in IoT device networks through a visual tool that utilizes three projection techniques, including Beta Hebbian Learning (BHL), t-distributed Stochastic Neighbor Embedding (t-SNE) and ISOMAP, in order to facilitate the identification of network attacks by human experts. This work research begins with the creation of a testing environment with IoT devices and web clients, simulating attacks over Message Queuing Telemetry Transport (MQTT) for recording all relevant traffic information. The unsupervised algorithms chosen provide a set (...) of projections that enable human experts to visually identify most attacks in real-time, making it a powerful tool that can be implemented in IoT environments easily. (shrink)

The present research describes a novel adaptive anomaly detection method to optimize the performance of nonlinear and time-varying systems. The proposal integrates a centroid-based approach with the real-time identification technique Recursive Least Squares. In order to find anomalies, the approach compares the present system dynamics with the average (centroid) of the dynamics found in earlier states for a given setpoint. The system labels the dynamics difference as an anomaly if it rises over a determinate threshold. To validate the proposal, two (...) different datasets obtained from a level control plant operation have been used, to which anomalies have been artificially added. The results shown have determined a satisfactory performance of the method, especially in those processes with low noise. (shrink)

Automatic control of physiological variables is one of the most active areas in biomedical engineering. This paper is centered in the prediction of the analgesic variables evolution in patients undergoing surgery. The proposal is based on the use of hybrid intelligent modelling methods. The study considers the Analgesia Nociception Index to assess the pain in the patient and remifentanil as intravenous analgesic. The model proposed is able to make a one-step-ahead prediction of the remifentanil dose corresponding to the current state (...) of the patient. The input information is the previous remifentanil dose, the ANI variable and the electromyogram signal. Modelling techniques used are Artificial Neural Networks and Support Vector machines for Regression combined with clustering methods. Both training and validation were done with a real dataset from different patients. Results obtained show the potential of this methodology to calculate the drug dose corresponding to a given analgesic state of the patient. (shrink)

Closed-loop administration of propofol for the control of hypnosis in anesthesia has evidenced an outperformance when comparing it with manual administration in terms of drug consumption and post-operative recovery of patients. Unlike other systems, the success of this strategy lies on the availability of a feedback variable capable of quantifying the current hypnotic state of the patient. However, the appearance of anomalies during the anesthetic process may result in inaccurate actions of the automatic controller. These anomalies may come from the (...) monitors, the syringe pumps, the actions of the surgeon or even from alterations in patients. This could produce adverse side effects that can affect the patient postoperative and reduce the safety of the patient in the operating room. Then, the use of anomaly detection techniques plays a significant role to avoid this undesirable situations. This work assesses different one-class intelligent techniques to detect anomalies in patients undergoing general anesthesia. Due to the difficulty of obtaining real data from anomaly situations, artificial outliers are generated to check the performance of each classifier. The final model presents successful performance. (shrink)

Batteries are a fundamental storage component due to its various applications in mobility, renewable energies and consumer electronics among others. Regardless of the battery typology, one key variable from a user’s perspective is the remaining energy in the battery. It is usually presented as the percentage of remaining energy compared to the total energy that can be stored and is labeled State Of Charge (SOC). This work addresses the development of a hybrid model based on a Lithium Iron Phosphate (LiFePO4) (...) power cell, due to its broad implementation. The proposed model calculates the SOC, by means of voltage and electric current as inputs and the latter as the output. Therefore, four models based on k-Means, Agglomerative Clustering, Gaussian Mixture and Spectral Clustering techniques have been tested in order to obtain an optimal solution. (shrink)

In this study, a hybrid model based on intelligent techniques is developed to predict the active power generated in a bioclimatic house by a low power wind turbine. Contrary to other researches that predict the generated power taking into account the speed and the direction of the wind, the model developed in this paper only uses the speed of the wind, measured mainly in a weather station from the government meteorological agency (MeteoGalicia). The wind speed is measured at different heights, (...) against the usual measurements in others researches, which uses the wind speed and the direction measured in a weather station on the wind turbine nacelle. The prediction is performed 30 minutes ahead, what ensures that the Building Management System knows the energy generated by the low power wind turbine 30 minutes before, and it can adapt the consumption of different equipment in the house to optimize the power use. The main objective is to allow the Building Management System to optimize the uses of energy, taking into account the predicted amount of energy that will be produced and the energy consumed in the house. The developed model uses a hybrid topology with four clusters to improve the prediction, achieving an error lower than 6.5% for Mean Absolute Error measured in a final test. To perform this test, part of the original dataset was isolated from the beginning of the training process to check the model with a dataset that is not used before, simulating the model as it is receiving new data. (shrink)