Alarm Prediction Research Articles

Overload Alarm Prediction in Power Distribution Transformers

The growing demand for electricity puts more strain on the grid, requiring automated and proactive strategies such as overload prediction to improve grid maintenance. However, the intermittent nature of power distribution loads makes the prediction more challenging. This paper proposes a novel framework for overload alarm prediction in distribution transformers, aimed at enhancing the reliability and efficiency of grid operations. Leveraging real-world smart meter data and machine learning techniques, the proposed system develops a classification model to predict overloads for distribution transformers. Due to resource constraints, a new strategy is adopted to assess the significance of alarms based on expert observations. Subsequently, a new approach is developed to imitate the experts, leading to an automated decision-making process using random forest. Ultimately, the transfer learning strategy is utilized to predict overload alarms for distribution transformers facing data scarcity in real-world applications. The proposed system demonstrates high accuracy of overload alarm predictions, paving the way for developing more proactive grid maintenance strategies.

APT: Alarm Prediction Transformer

Distributed control systems (DCS) are essential to operate complex industrial processes. A major part of a DCS is the alarm system, which helps plant operators to keep the processes stable and safe. Alarms are defined as threshold values on individual signals taking into account minimum reaction time of the human operator. In reality, however, alarms are often noisy and overwhelming, and thus can be easily overlooked by the operators. Early alarm prediction can give the operator more time to react and introduce corrective actions to avoid downtime and negative impact on human safety and the environment. In this context, we introduce Alarm Prediction Transformer (APT), a multimodal Transformer-based machine learning model for early alarm prediction based on the combination of recent events and signal data. Specifically, we propose two novel fusion strategies and three methods of label encoding with various levels of granularity. Given a window of several minutes of event logs and signal data, our model predicts whether an alarm is going to be triggered after a few minutes and, if yes, it also predicts its location. Our experiments on two novel real industrial plant data sets and a simulated data set show that the model is capable of predicting alarms with the given horizon and that our proposed fusion technique combining inputs from different modalities, i. e. events and signals, yields more accurate results than any of the modalities alone or conventional fusion techniques.

PREDICTING FIRE ALARMS USING MULTI SENSOR DATA: A BINARY CLASSIFICATION APPROACH

Fires pose significant threats to human life, property, and the environment. Early detection of fire incidents is crucial to prevent extensive damage and to ensure the safety of occupants. Traditional fire alarm systems typically rely on a single type of sensor, such as smoke detectors or heat sensors, to detect specific fire indicators. These systems operate based on predefined thresholds and triggers. However, they can be prone to false alarms triggered by non-fire-related events (e.g., cooking fumes or dust) and may not provide early warning signs in certain scenarios. To address these limitations, researchers and engineers have turned to advanced technologies, such as multi-sensor data analysis and machine learning algorithms, to develop more reliable and efficient fire alarm prediction systems. On the other hand, the need for a more robust and accurate fire alarm prediction system stems from the shortcomings of traditional methods. False alarms not only lead to wasted resources but also desensitize occupants, potentially leading them to ignore genuine alarms. Additionally, a delayed response to a fire incident can result in severe consequences, making it essential to develop an intelligent system that can effectively and timely predict fire events. Therefore, this work presents the utilization of multi-sensor data and binary classification to develop a more reliable fire alarm prediction system. The experiments are conducted using a dataset collected from various sensor inputs, including air temperature, humidity, CO2 concentration, molecular hydrogen, ethanol gas, and air pressure etc. Then applied binary classification algorithm to learn patterns from the data and classify fire-related events accurately. The results showed promising improvements in prediction accuracy, reduced false alarm rates, and early detection of fire incidents.

Structure Feature Extraction for Hierarchical Alarm Flood Classification and Alarm Prediction

Structure Feature Extraction for Hierarchical Alarm Flood Classification and Alarm Prediction

GTFP: Network Fault Prediction Based on Graph and Time Series

With the explosion of 5G network scale, the network structure becomes increasingly complex. During the operation of the network devices, the probability of anomalies or faults increases accordingly. Network faults may lead to the disappearance of important information and cause unpredictable losses. The prediction of network faults can enhance the quality of network services and reduce economic loss. In this paper, we propose the concept of 4D features and use the BERT algorithm to extract semantic features, the graph neural network algorithm to extract network topology information, and the Temporal Convolutional Network (TCN) algorithm to extract time series. Based on this, we propose Fault Prediction based on GraphSage and TCN (GTFP), an end-to-end solution of network fault alarm prediction, which is based on GraphSage and TCN (GTCN), a hybrid algorithm of a graph neural network and the TCN model. Our solution takes the historical alarm data as input. First, we filter out the alarm noises irrelevant to the faults through data cleaning. Then, we employ feature engineering to extract the valid alarm features, including the statistical features of the network alarm information, the semantic features of the alarm texts, the sequential features of the alarms and the network topology features of the nodes where the alarms are located. Finally, we use GTCN to predict future fault alarms based on the extracted features. Experiments on the alarm data of a real service system show that GTFP performs better than the state-of-the-art algorithms of fault alarm prediction.

A priority-aware sequential pattern mining method for detection of compact patterns from alarm floods

Alarm floods are commonly present in modern complex industrial process, and usually degrade the performance of the alarm systems severely. In an alarm flood, the massive alarm messages may overwhelm plant operators and submerge critical alarms, which would further worsen the situation and lead to severe consequences. One effective way to deal with alarm flood situations is to mine meaningful alarm patterns that can be used to configure dynamic alarm suppression modules that suppress irrelevant alarms in alarm floods based on certain conditions or to help with online alarm prediction. Using the existing methods in literature, alarms of higher priorities could be easily overlooked in the pattern mining as they usually appear less frequent; thus, it makes the results become unreliable. Motivated by this issue, this paper proposes a new pattern mining method to extract priority-aware compact alarm patterns from historical alarm flood sequences. The main contributions of this study are twofold: (1) a priority-aware sequential pattern mining algorithm is designed to mine alarm sequential patterns using a closed sequential pattern mining algorithm with alarm priority information incorporated; (2) a pattern compression strategy is proposed to merge similar sequences into compact patterns, making it easier to search for alarm patterns. The effectiveness of the proposed pattern mining method is demonstrated by alarm data generated using a vinyl acetate monomer chemical plant model.

An association rule mining approach to predict alarm events in industrial alarm floods

Industrial operators often experience overwhelming situations during ongoing alarm floods due to high alarm rates. In such situations, real-time assistance in the form of prediction of upcoming alarm events can ease off the decision-making for industrial operators. Accordingly, this work studies alarm prediction in alarm flood situations, and the main contribution lies in a novel association rule mining approach for real-time prediction of alarm events and their corresponding times of annunciation during an ongoing alarm flood. The proposed method is capable of performing predictions at the triggering instant and modifying the predictions with the increasing of the ongoing alarm flood. The proposed method is implemented mainly in the following steps: (1) A Compact Prediction Tree (CPT) model is modified with new features, namely, the time table and co-occurrence matrix, and constructed based on historical alarm sequences; (2) an alarm relevancy detection strategy is designed to detect and eliminate irrelevant alarms in alarm floods; (3) an online alarm prediction algorithm is designed to predict upcoming alarms at the early stage of the ongoing alarm flood; (4) the confidence intervals of the time differences between the annunciations of subsequent predicted alarm events are calculated for time prediction. To demonstrate the effectiveness of the proposed method, an industrial case study based on real alarm & event logs from an oil refinery is provided.

Sequence Compression and Alignment-Based Process Alarm Prediction

Sequence Compression and Alignment-Based Process Alarm Prediction

Adaptive alarm prediction in optical network based on model generalization in cross-layer AI

Artificial intelligence (AI) applications about optical network management are increasing, especially in fault diagnosis. When the network fault or alarm occurs, the equipment running status and performance parameters are changing. AI algorithms can extract the internal relationship from the data and it is proved that the alarm prediction method based on AI is effective. However, the performance of AI models, especially supervised learning, rely on large data sets. These training sets with only a small number of samples are usually unavailable due to the lack of historical data. How to improve the generalization ability of AI models is still a key problem to be solved in the intelligent maintenance of optical networks. In optical network alarm prediction, performance data of devices have different degrees of relationship to alarms and the generalization performance of models is limited by the data structure. In this paper, we propose a method to optimize the model generalization ability in alarm prediction, i.e., the training of a small sample model is solved through parameter transfer in the training process. We verify the generalization effect of the models by using real data of a commercial optical network. Experimental results show that the models achieve significant performance improvement in the data set with a high correlation between performance and alarm.

Robust deep learning early alarm prediction model based on the behavioural smell for android malware

Due to the widespread expansion of the Android malware industry, malicious Android processes mining became a necessity to understand their behavior. Nevertheless, due to the complexities of size, length, and associations of some essential and distinguishing Android applications’ features such as API calls and system calls, mining malicious Android processes become a prominent obstacle. The malicious process mining obstacle is also coupled with the increasing rate of zero-day attacks, with no prior knowledge about those kinds of behaviors. Hence, malware detection alone is no longer enough; instead, we need new methodologies to predict malicious behaviors early. In this paper, we propose a behavioral Android malware smell predictor model. Our model relies on various static and dynamic features. We overcame the problem of massive feature size and complex associations by encapsulating related features in a few cluster classes. Accordingly, the cluster classes are exchangeably used to represent the features in the original calling sequences. Regarding substantially long sequences, experimental results showed that our model could predict whether a process is behaving maliciously or not based on rapid-sequence-snapshot analysis. The proposed model counted on the LSTM model to classify the reformed API and system call sequences snapshots. Moreover, we used ensemble machine learning classifiers to classify Android permissions. We trained the LSTM model using random snapshots of the newly formed API and system call cluster sequences. We tested our model against common ransomware attacks. We found that our trained LSTM model showed stable performance at a particular snapshot size. The model showed competitive accuracy in predicting new sequences. Accordingly, we proposed an early alarm solution for blocking malicious payloads instead of identifying them after their fulfillment. Hence, we can avoid the cost of future damage.

Automation of Photonic Networks Using Machine Learning: Case Studies and Future Works

Although a “Self-Driving” photonic network is still a long way to go, many time-consuming complex tasks and decision making in photonic networks can be automated using machine learning, and other data-driven solutions. This study explores recent contributions towards photonic network automation, such as alarm prediction, fault localization, resource auto-scaling, quality of transmission prediction, dynamic controller placement, automated service restoration, resource allocation and optimization, minimization of electricity and power supply cost, user data analysis, etc. The studies explored in this letter, provide solution approaches to these problems using wide range of data-driven methods, machine learning, AI, and deep learning based methods. This study discusses different challenges involving these interesting research areas and provides directions for future research opportunities as well.

A maximum-entropy-based method for alarm flood prediction

Alarm floods are the phenomena of presenting too many alarms in a short time period to exceed abilities of industrial plant operators in making proper responses. This paper proposes a maximum-entropy-based method to predict upcoming alarms for an occurring alarm flood. The proposed method has two important features: all currently-occurred alarms are exploited for prediction, and upcoming alarms are given with quantitative probabilities. By contrast, existing alarm prediction methods either use most-recent (not all) occurred alarms in prediction, or cannot predict specific alarms with quantitative probabilistic values. The proposed method takes all historical alarm flood sequences into account to establish relationships between currently-occurred alarms and upcoming alarms, and formulates an optimization problem to maximize conditional entropies of upcoming alarms. The effectiveness of the proposed method is validated by numerical examples, one of which is on the well-accepted benchmark of Tennessee Eastman process.

Alarm Prediction in Cellular Base Stations Using Data-Driven Methods

The importance of cellular networks continuously increases as we assume ubiquitous connectivity in our daily lives. As a result, the underlying core telecom systems have very high reliability and availability requirements, that are sometimes hard to meet. This study presents a proactive approach that could aid satisfying these high requirements on reliability and availability by predicting future base station alarms. A data set containing 231 internal performance measures from cellular (4G) base stations is correlated with a data set containing base station alarms. Next, two experiments are used to investigate (i) the alarm prediction performance of six machine learning models, and (ii) how different predict-ahead times (ranging from 10 min to 48 hours) affect the predictive performance. A 10-fold cross validation evaluation approach and statistical analysis suggested that the Random Forest models showed best performance. Further, the results indicate the feasibility of predicting severe alarms one hour in advance with a precision of 0.812 (±0.022, 95 % CI), recall of 0.619 (±0.027) and F1-score of 0.702 (±0.022). A model interpretation package, ELI5, was used to identify the most influential features in order to gain model insight. Overall, the results are promising and indicate the potential of an early-warning system that enables a proactive means for achieving high reliability and availability requirements.

Comparison between ANN and random forest for leakage current alarm prediction

In order to improve the efficiency of the electrical safety operations of private electric facilities, the use of AI and IoT is expected. In this paper, we propose a leakage current alarm prediction model using a random forest and an artificial neural network. Customer information, periodic inspection history, alarm occasions on the previous day, and weather information are used as explanatory variables. A grid search was performed for hyperparameter optimization of each model, and generalization performance was evaluated using OOB verification and cross-validation. As a result of comparing the performances of the two models by the PR curve, it was found that the random forest had a larger PR curve and had better prediction performance.

A flexible alarm prediction system for smart manufacturing scenarios following a forecaster–analyzer approach

The introduction of data-related information technologies in manufacturing allows to capture large volumes of data from the sensors monitoring the production processes and different alarms associated to them. An early prediction of those alarms can bring several benefits to manufacturing companies such as predictive maintenance of the equipment, or production optimization. This paper introduces a new system that allows to anticipate the activation of several alarms and thus, warns the operators in the plants about situations that could hamper the machines operation or stop the production process. The system follows a two-stage forecaster–analyzer approach on which first, a long short-term memory recurrent neural network based forecaster predicts the future sensor’s measurements and then, distinct analyzers based on residual neural networks determine whether the predicted measurements will trigger an alarm or not. The system supports some features that make it particularly suitable for smart manufacturing scenarios: on the one hand, the forecaster is able to predict the future measurements of different types of time-series data captured by various sensors in non-stationary environments with dynamically changing processes. On the other hand, the analyzers are able to detect alarms that can be modeled with simple rules based on the activation condition, and also more complex alarms on which it is unknown when the activation condition will be fulfilled. Moreover, the followed approach for building the system makes it flexible and extensible for other predictive analysis tasks. The system has shown a great performance to predict three different types of alarms.

Development of Alarm Prediction System for Monitoring Steam Turbine Based on SCADA Data

Development of Alarm Prediction System for Monitoring Steam Turbine Based on SCADA Data

Alarm classification prediction based on cross-layer artificial intelligence interaction in self-optimized optical networks (SOON)

Abstract Alarm prediction in optical networks focuses on forecasting network failure from the state of equipment and links. The existing prediction methods usually rely on large amounts of data, while centralizing all processes in the network controller or management system may increase the system burden. In this paper, a novel method is proposed in self-optimized optical networks (SOON) to implement alarm classification prediction based on cross-layer artificial intelligence (AI) architecture. We adopts alarm risk assessment and data augmentation with synthetic minority oversampling technique (SMOTE). As a distributed system, cross-layer AI completes decomposed functions by using interactions between different AI engines. With the help of the controller system, functions can be executed in order. The amount of data required for prediction is far less than other methods. The validity of the method is proved using the collected data from a commercial synchronous digital hierarchy (SDH) network. Experimental results show that promising precision (95%) can be achieved in predicting the optical equipment alarms.

Aiding Alarm Rationalization by Automatic Identification of various sequential patterns in large volume of Alarm and Event log data

The main problem with modern Industrial Alarm System is that the operators receive far too many alarms than they can handle. Many of the alarms presented to the operators are nuisance in nature such as chattering alarms, redundant alarms, consequential alarms, and alarms with no action. Such alarms superfluously increase the alarm load to an operator, do not essentially indicate any real abnormal situation and more importantly do not require any corrective actions by the operator. Being overwhelmed by large number of nuisance alarms, operators always face tough challenge in identifying the real alarms which demands prompt operator actions and are often buried in the pool of nuisance alarms. It is therefore necessary to reduce the alarms loads of an operator by systematically identifying the nuisance alarms and presenting only the real useful alarms to the operators. Also, there are some alarms which require well defined operator actions to bring them back to normal. When these alarms get triggered operators are supposed to take a well-defined set of corrective action in order to normalize them. Identification of these alarms for which there is a consistent operator action would be of immense help since operators can immediately focus their attention on these alarms whenever they occur and carryout the necessary actions. The main goal of Alarm rationalization is to minimize alarm load to the operators by presenting only the alarms that are relevant and require operator actions. Alarm Rationalization process by and large is carried out manually. It is quite tedious, time consuming and requires a lot of manual efforts. To solve the these challenges and to aid in Alarm Rationalization process, a novel approach has been proposed in this paper for identifying the nuisance alarms by analyzing the large volume of Alarm and Event log data. Several Alarm Rationalization activities (such as: Changing Alarm Configuration settings, Alarm Suppression, Alarm grouping, Alarm removal, Alarm Prediction, Standardization of Alarm response Procedure) can be performed by based on the identified chattering alarms. The proposed approach is illustrated through some examples obtained by analysing the Alarm and Event log data from petrochemical and water treatment plants in Singapore.

Power Environment Warning Prediction Model Based on Big Data Association Rules

With the rapid development of power communication network, power production and operation put forward higher requirements for the stability of power communication network. The power environment system can collect the operation status of communication power supply, equipment and other related alarm information as the basis for troubleshooting. However, due to the failure of a device, there will be some columns of linkage alarm, which increases the difficulty of fault location. In order to solve the above problems, this paper uses the method of big data association rules to carry out data deep mining, and makes in-depth analysis on the location and prediction of power environment fault alarms. Through the correlation between power environment alarms, this paper puts forward the analysis and prediction method of power environment alarms, and realizes the prediction of power environment alarms through the analysis of big data association rules. Based on the in-depth analysis of the Apriori algorithm, this paper designed an improved Apriori algorithm which is suitable for power environment alarm prediction, and greatly improves the efficiency of the algorithm.

Optimal hyperparameters for random forest to predict leakage current alarm on premises

While the number of private electrical facilities is increasing, there are not enough security personnel to perform the security work. In this paper, we propose a random forest model for predicting leakage current alarms in order to improve the efficiency of electrical safety operations. A random forest was created using periodic inspection data, alarm data, and meteorological data as explanatory variables, and generalization performance was evaluated by OOB-based F-measure. In order to obtain the highest performance, a grid search was performed to optimize the hyperparameters. As a result, it was possible to achieve alarm prediction with a certain level of performance. In addition, the optimal hyperparameters were found by grid search, and the F-measure was improved.