In recent times, cyber security offers a significant advancement in smart grid technologies for its availability and functionality. The potential intrusion in smart grids marks the system to behave in a vulnerable way all the private data. Smart grids are often prone to data integrity attacks at its physical layer, which is been a critical issue presently. This attack alters the measurement of compromised meter set by the attacker(s). It misleads the decision making by the operators at the control center and thereby the reliability of the measurement is affected. In this chapter, the authors present a deep learning ensemble (DLE) model that possibly detects the potential data integrity attacks in the physical layer. The deep learning model uses ensemble learning to make decisions and combines the classified results to improve the classification on test data. The experiments are conducted on the proposed DLE model to find the accuracy of classifier the malicious and benign measurements.
TopIntroduction
In the modern era, the integration of power systems with communication technologies has improved the efficiency, reliability and consumption of electrical energy. The enabling of Advanced Metering Infrastructure (AMI) in smart grids offers transparency and reduces the consumption. However, it opens us the possibility of attacks in the form of cyber threats, where the entire system and it communication medium is highly prone to cyber-attacks .(Khanna, K.,et al., 2016).
These intrusions makes the smart grids susceptible to attacks that leads to serious degradation like leading of private information and system failures (Basodi, S.,et al., 2020). It further targets the confidentiality, data delivery and integrity, which may lead to financial, loses through theft, power grid instability or poor accessing of critical data (Farraj, A.,et al.,2017). In future, the Smart Grid may incorporate various other technologies and innovations like sensing, communication and distributed control for the accommodation of Electric Vehicle loads, renewable generation and storage (Giani, A.,et al., 2013). Hence, cyber security offers a significant role on smart grids by determining the possible potential intrusions.
In this study, we consider the attacks in smart grids occur due to the data integrity attack that alters the compromised power meter readings. The unobservable attack (power flow constraints on consistent compromised meter readings). A proper coordination is required by the unobservable attacks on the compromised meter readings and it should be orchestrated in a careful manner. Such that it operates on a low dimensional manifold to ensure that the attack is unobservable and it misleads the system operators by offering substantial errors in the state estimation algorithms. The vulnerability of grids is more prone to cyber-attacks, thus increasing the relevance and urgency on cyber security research (Giani, A.,et al., 2013).
There exist various efforts to analyze the data integrity attacks in smart grid environment that develops alternative strategies such that the damages can be reduced temporarily. The methods developed are accurate, quick and offers the detection in a cost-effective manner to mitigate the data integrity attacks and ensures improved security on smart grids(Ge, L.,et al., 2017).
Operation of smart grids under uncertainties leads to cyber-attacks, failures, poor quality of service, error in device synchronization, compromising the resources capacity, and so on. In addition, with increased and rapid communication across the communication channel, the existing methods tend to fail in processing large data while data integrity attack is on surge. Diagnosing these challenges finds that there are numerous concerns associated with smart grids that compromise on its resilience and security. To mitigate such challenges, we develop a deep learning ensemble classification model with a meta-heuristic feature extraction that process effectively the data in a larger scale basis and offers effective computations. The main contributions are stated below:
- a)
The author has developed a malicious classifier model that forms a series of framework including data pre-processing and normalisation, feature extraction using bee swarm optimisation (BSO) and ensemble classification.
- b)
The ensemble classifier is designed with a series of machine learning base classifier namely the neural network and the results of base classes are combined using ensemble technique and finally sent as input for deep neural network (DNN) classifiers to detect the malicious data samples from the input datasets.
- c)
The input datasets are generated in massive on the IEEE bus systems including 30-bus and 57-bus and 118-bus systems. 70% of the input dataset is used as training data and the remaining data is used as testing datasets. The performance of the entire system under all the three bus systems is compared with existing classifiers against various performance metrics.
The outline of the paper is presented below: Section 2 offers the related works in the given field. Section 3 discusses the model and the deep learning classification model. Section 4 provides the dataset and section 5 provides the results and discussion of the present study. Finally, the section 6 concludes the entire work.