**1. Introduction**

The Cognitive Dynamic System (CDS) is an organized physical model and research tool that is based on certain features of the brain. Following its first introduction in [1], it was later expanded in [2] leading to its first applications in cognitive radio [3] and cognitive radar [4]. Since then, CDS has progressed enormously to give rise to Cognitive Control (CC) [5] and Cognitive Risk Control (CRC) [6] as two of its particular functions. Using those principles, the CDS was first merged in [7] with the Smart Grid (SG) to form a new structure, based on the DC state estimation model, that shows tremendous potential for handling the possible problems that the SG will be facing in the near future. Furthermore, in [8], the construct presented in [7] was expanded to include a more complex CRC that is closer to the brain. In that paper, it was proven how this new approach can to be used to mitigate the problem of cyber-attack in the SG. From a neuroscience perspective, the CDS is founded on Fuster's paradigm of cognition comprising of the following five principles: perception-action cycle (PAC), memory, attention,

intelligence and language [9]. In its simplest form, the CDS is built on two main components: the perceptor, on one side, and the executive on the other with the feedback channel uniting them together. In [7], it was shown that the integration of the over-arching function of CDS, CC, with the SG, is well adapted for slowly progressing cyber-physical systems. In this chapter, the construct presented in [7], where the DC-estimation model was involved, will be re-engineered to be able to carry out AC state estimation optimally and also be able to detect cyber-attacks. In order to do so, the perceptor of the CDS will incorporate a generative model that will allow it to sense and control the environment indirectly. Moreover, in order to bring forward the cognitive ability of the CDS and make it compatible with the current nonlinear state estimation in SG, the steps involved in the state estimation process will be re-engineered in a novel way. It will also be shown how the entropic state, which is the objective function of the CDS, will be instrumental in implementing a control-sensing mechanism that is capable of identifying and handling bad measurements. We will also show how this entropic state serves as the basis for detecting False Data Injection attacks (FDI) in SG.

#### **1.1 Smart grid**

The next generation of engineering systems consisting of the Internet of Things (IoT) and Cyber-physical systems (CPSs) are currently paving the way towards the fourth industrial revolution [10]. As those systems are gradually occupying a more prominent role in our daily lives, through applications in critical infrastructures such as electrical power grids or transportation systems, the cyber-security aspects of those systems will also grow in importance [11]. In the context of this chapter, emphasis will be laid upon the SG and its most dangerous threat known as False Data Injection (FDI) attacks. More specifically, compared to our previous research where the DC model for state estimation was investigated [7], focus will be laid upon on the AC model, which is more a realistic representation of the smart grid, and the introduction the CDS for a new way of control and FDI attack detection.

Making use of all the new generation of sensing, monitoring and control strategies, the SG is forecasted to be a more powerful entity than the traditional power grid in many facets such as reliability and efficiency [12, 13]. In the SG, the Supervisory Control and Data Acquisition systems (SCADA) is responsible for monitoring and processing the main control actions by collecting meter measurements from remote terminal units (RTUs) consisting of different field devices or sensors. Through a process known as state estimation, those measurements are then processed and analyzed for errors and inconsistencies after being transmitted to a control center [14, 15]. The state variables that are calculated by this process usually consist of the voltage magnitudes and angles of the different busses in the system [16]. The measurements used for state estimation are the currents, real and reactive power flows, power injections and voltage magnitudes and angles. In the DC model, the state variables are the bus angles only while in the more complex AC model, the voltage magnitudes and angles of the different busses in the network are estimated. Weighted Least Squares (WLS), introduced by Schweppe [14], is the technique used for the power system state estimation using those measurements. In order to enhance the accuracy of the estimated states, another process, known as Bad Data Identification, is carried out to remove bad measurements. Bad measurements are erroneous measurement readings that will impact state estimation negatively. The most commonly applied bad data identification techniques are the Chi-Squared Tests and Largest Normalized Residual Test [15, 17]. Those statistical tests rely on the residuals between the estimated states and the measurement residuals to identify the bad data. In the case of an FDI attack, bad data, which can bypass the

*Cognitive Dynamic System for AC State Estimation and Cyber-Attack Detection in Smart Grid DOI: http://dx.doi.org/10.5772/intechopen.94093*

previously mentioned tests, is introduced into the system such that the estimated states can be modified stealthily. Those bad data are maliciously crafted offsets to measurements that are injected to the sensor readings so as the state estimation process is influenced in a particular way. Consequently, with the incorrect calculated states, bad control decisions will be applied.

Although FDI attacks have been a popular topic of research over the past years [18], most of the works, e.g., in [10–13, 19], investigated the FDI attacks on the DC model. Few works have been published on the AC model and those attacks [18, 20, 21]. Nevertheless, the DC model is just a simplified representation of the nonlinear AC state estimation model. There are major differences between the two models that could explain why the AC model has been unpopular. Firstly, in the nonlinear state estimation model, the estimated states are obtained after undergoing iterations, while in the DC model, those states are obtained in closed-form. Moreover, the linear state estimation relies on active power flow analysis [16, 22, 23]. On the other hand, the AC model uses both active and reactive power flow analysis. Furthermore, the state variables in the DC model consist of the voltage angles only while the states in the AC model consist of both the voltage angles and magnitudes. Consequently, these differences raise the complexity and computational expense of nonlinear state estimation as a topic of research when it comes to FDI attacks [24]. In fact, DC based FDI attacks can be detected by AC-based data detection techniques [20]. Hence, since the AC model is commonly applied in power systems, finding a way to detect these attacks and mitigating them under that environment is going to be very important for the coming years.

## **1.2 Contribution and organization**

The main contributions of this chapter can be summarized as follows:


The rest of this chapter is organized as follows: In Section 2, the basic concepts of state estimation and data detection for the AC model will be presented and contrasted. The mathematics of FDI attacks for this model will also be demonstrated. Section 3 expands on the structure of the CDS for the SG. Since this research is an extension of [7], the material presented in that paper will be re-engineered for this new application. In the context of the CDS, the SG is considered as the environment with which it interacts. Section 4 gives a discussion on the application and simulation results of this approach on the IEEE 14-bus network. It will be shown how this new structure is able to handle the two problems of bad data detection and FDI attack detection simultaneously. Finally, Section 5 concludes this paper by highlighting the key results and presenting new avenues of research for this novel construct.
