Anomaly detection in data storage systems is a challenging problem due to the high dimensional sequential data involved, and lack of labels. The state of the art for automating anomaly detection in these systems typically relies on hand crafted rules and thresholds which mainly allow to distinguish between normal and abnormal behavior of each indicator in isolation. In this work we present an end-to-end framework based on convolutional autoencoders which not only allows for anomaly detection on multivariate time series data, but also provides explainability. This is done by identifying similar historic anomalies and extracting the most influential indicators. These are then presented to relevant personnel such as system designers and architects, or to support engineers for further analysis. We demonstrate the application of this framework along with an intuitive interactive web interface which was developed for data storage system anomaly detection. We discuss how this framework along with its explainability aspects enables support engineers to effectively tackle abnormal behaviors, all while allowing for crucial feedback.

Detecting anomalies from high-dimensional multivariate temporal data is challenging, because of the non-linear, complex relationships between signals. Recently, deep learning methods based on autoencoders have been shown to capture these relationships and accurately discern between normal and abnormal patterns of behavior, even in fully unsupervised scenarios. However, validating the anomalies detected is difficult without additional explanations. In this paper, we extend SHAP -- a unified framework for providing additive explanations, previously applied for supervised models -- with influence weighting, in order to explain anomalies detected from multivariate time series with a GRU-based autoencoder. Namely, we extract the signals that contribute most to an anomaly and those that counteract it. We evaluate our approach on two use cases and show that we can generate insightful explanations for both single and multiple anomalies.

In this work we present MTEX-CNN, a novel explainable convolutional neural network architecture which can not only be used for making predictions based on multivariate time series data, but also for explaining these predictions. The network architecture consists of two stages and utilizes particular kernel sizes. This allows us to apply gradient based methods for generating saliency maps for both the time dimension and the features. The first stage of the architecture explains which features are most significant to the predictions, while the second stage explains which time segments are the most significant. We validate our approach on two use cases, namely to predict rare server outages in the wild, as well as the average energy production of photovoltaic power plants based on a benchmark data set. We show that our explanations shed light over what the model has learned. We validate this by retraining the network using the most significant features extracted from the explanations and retaining similar performance to training with the full set of features.

We demonstrate that CNN deep neural networks can not only be used for making predictions based on multivariate time series data, but also for explaining these predictions. This is important for a number of applications where predictions are the basis for decisions and actions. Hence, confidence in the prediction result is crucial. We design a two stage convolutional neural network architecture which uses particular kernel sizes. This allows us to utilise gradient based techniques for generating saliency maps for both the time dimension and the features. These are then used for explaining which features during which time interval are responsible for a given prediction, as well as explaining during which time intervals was the joint contribution of all features most important for that prediction. We demonstrate our approach for predicting the average energy production of photovoltaic power plants and for explaining these predictions.

Today's operation of buildings is either based on simple dashboards that are not scalable to thousands of sensor data or on rules that provide very limited fault information only. In either case considerable manual effort is required for diagnosing building operation problems related to energy usage or occupant comfort. We present a Cognitive Building demo that uses (i) semantic reasoning to model physical relationships of sensors and systems, (ii) machine learning to predict and detect anomalies in energy flow, occupancy and user comfort, and (iii) speech-enabled Augmented Reality interfaces for immersive interaction with thousands of devices. Our demo analyzes data from more than 3,300 sensors and shows how we can automatically diagnose building operation problems.

The detection and diagnosis of abnormal building behavior is key to further improve the comfort and energy efficiency in buildings. An increasing number of sensors can be utilized for this task but these lead to higher integration effort and the need to capture the sensor interactions. This paper presents a novel diagnostic approach for buildings with complex heating, ventilation, air-conditioning (HVAC) systems. It uses semantic graphs to automatically create the diagnostic model from the building's data points and to identify potential cause-effect-relationships based on past and current time series data. The approach is validated on various simulated examples of a multiroom HVAC control system. The experimental results show that it can diagnose multiple faults with and without delays with high accuracy.

Many various types of sensors coming from different complex devices collect data from a city. Their underlying data representation follows specific manufacturer specifications that have possibly incomplete descriptions (in ontology) alignments. This paper addresses the problem of determining accurate and complete matching of ontologies given some common descriptions and their pre-determined high level alignments. In this context the problem of ontology matching consists of automatically determining all matching given the latter alignments, and manually verifying the matching results. Especially for applications where it is crucial that ontologies are matched correctly the latter can turn into a very time-consuming task for the user. This paper tackles this challenge and addresses the problem of computing the minimum number of user inputs needed to verify all matchings. We show how to represent this problem as a reasoning problem over a bipartite graph and how to encode it over pseudo Boolean constraints. Experiments show that our approach can be successfully applied to real-world data sets.

Understanding the operation of a building is key for improving it and reducing energy waste. However, today this is a mostly manual task for which domain experts use visual tools to analyze the large amounts of building data. We show how to automate this task by means of pattern extraction techniques. These allow human operators to simply consider well defined data patterns rather than vast amounts of data. Here the manual effort consists in identifying the context in which the patterns occur. In this paper we go one step further and show how we can automatically derive also the context in which building operation patterns occur by considering the influencing factor that govern building operation. We have evaluated our approach for two pattern extraction methods: matrix factorization and clustering. Our experimental results using real world data demonstrate the applicability of our work.

Sensor systems are constantly growing in all application areas and become elements of our environment. Semantic Sensor Networks (SSN) support this development and provide standardized semantic access for reasoning on this information. Unfortunately they do not model internal system knowledge or simple correlations between sensors and hence they cannot be used to automatically perform analytics tasks based on sensor data only. We show how SSN ontology can be extended and demonstrate its benefits for the task of diagnosing smart building problems using real-world data.

Diagnosis is the task of explaining abnormal behaviors of systems like telecommunication, transportation or energy systems. Given a sequence of observations the problem is to determine, online, all faults that are in line with these observations. Many approaches tackle this problem but they either require domain expertise or a formal description of how observations and faults are connected. This limits their scope to the diagnosis of well-understood faults. We address the problem of diagnosing faults that may occur for the first time and present a new diagnosis approach that integrates techniques for analyzing semantic descriptions of observations and faults.

The timely detection of abnormal energy usage is one of the major ad-hoc techniques to optimize energy efficiency. Typically an alarm is triggered either by a significant drift from the baseline consumption level or by a period of large variations. In this paper we propose a statistical predictive method for detecting anomalies both in mean and in variation. The criterion behind is based on the prediction intervals (PIs) of the baseline, which is estimated by the Generalized Additive Model (GAM), and of the variations of baseline, which is estimated by the Autoregressive Conditional Heteroscedastic Model (ARCH). Our proposal on systematically studying the time-dependent variations of energy consumption by ARCH is novel. This is of great importance to, technically, guarantee the resulting PIs of baseline is valid and, practically, to reduce the energy cost incurred by oscillation. As a key component of anomaly detection algorithm, we propose to use the residual based bootstrap for the construction of PIs to minimize the bias caused by imposing hypothetical distributions on observations. We illustrate the proposed method with a real-life example on building energy consumption throughout the paper and in addition, justify our approach is theoretically consistent.

The goal of testing is to discriminate between multiple hypotheses about a system - for example, different fault diagnoses - by applying input patterns and verifying or falsifying the hypotheses from the observed outputs. Definitely discriminating tests (DDTs) are those input patterns that are guaranteed to discriminate between different hypotheses of non-deterministic systems. Finding DDTs is important in practice, but can be very expensive. Even more challenging is the problem of finding a DDT that minimizes the cost of the testing process, i.e., an input pattern that can be most cheaply enforced and that is a DDT. This paper addresses both problems. We show how we can transform a given problem into a Boolean structure in decomposable negation normal form (DNNF), and extract from it a Boolean formula whose models correspond to DDTs. This allows us to harness recent advances in both knowledge compilation and satisfiability for efficient and scalable DDT computation in practice. Furthermore, we show how we can generate a DNNF structure compactly encoding all DDTs of the problem and use it to obtain a cost-optimal DDT in time linear in the size of the structure. Experimental results from a real-world application show that our method can compute DDTs in less than 1 second for instances that were previously intractable, and cost-optimal DDTs in less than 20 seconds where previous approaches could not even compute an arbitrary DDT.

Automated fault diagnosis has significant practical impact by improving reliability and facilitating maintenance of systems [1]. Given a monitor continuously receiving observations from a dynamic eventdriven system, diagnosis algorithms infer possible fault events that