Applications of Artificial Intelligence in Planning and Operation of Smart Grids

The electricity market operation in smart grid requires a certain level of intelligence by the agents. The operation of multiple agents in the market is different under various scenarios where the machine learning approach exhibited by the agents plays a major role. Each trader is focused on achieving high profit yielding and a positive reward by experimenting various cost parameters and propensity levels. The best learning model for smart grid incorporating the deregulated market structure is reinforcement learning. In this type of model, the agent takes action in an environment based on the previous experience and improvises automatically in the successive bidding environment. This help the generator companies to maximize their profit and achieve smart bidding environment for efficient smart grid transactions. Once the generator fails to meet the requirement and propose greater price, penalization also becomes a part to assess the bidding strategy in electricity market through agent-based machine learning. A new interactive Variant Roth-Erev algorithm based approach is implemented in various agents, and the test system considered for the entire analysis is an IEEE 5-bus system. The system is modelled taking into account different propensity levels and stopping rules. Congestion handling in the system is also an objective irrespective of earning high profit. Further, the analysis is extended for a period of 50 days to showcase the effectiveness of the approach. The results obtained for different cases are investigated in detailed way through comparison without learning and with machine learning approach using Java-based programming.

Photovoltaic (PV) systems have become a potential solution to global problems like pollution and climate changes resulting from the excessive use of fossil fuels. This kind of system can respond to the constant increase in the electric energy demand and the need for energy supply in rural or hard-to-reach areas. However, as the energy efficiency of PV systems is low, there exists a necessity to maximize the output power so that it reaches the maximum power point (MPP). Different Maximum Power Point Tracking (MPPT) techniques can be used to increase the efficiency of PV systems. Nevertheless, climatic variations make their task difficult to achieve. This work proposes the use of reinforcement learning (RL) techniques for solving the MPPT problem of a PV system under different conditions of temperature and solar irradiance. RL techniques do not require information of a model that describes the behavior of the system with its environment. They only make use of the information of the possible states to visit and actions to take and updates a utility function according to how good the last action taken was. To validate the effectiveness of the proposed algorithms, several experiments were performed in a simulated environment. The obtained results show good performances with stable behaviors, proving to be practical for the control of photovoltaic systems.

Parameters of solar radiation are unstable, which affect the prediction accuracy of photovoltaic (PV) power. Many hybrid models have been applied recently to improve the prediction accuracy of the models. In this chapter, a novel three-stage hybrid model of Neighborhood Component Analysis (NCA), and Artificial Neuronal Network (ANN) optimized with Particle Swarm Optimization (PSO) is introduced. First, the NCA technique is applied as a feature extraction technique to determine the relevant features that have substantial influence on photovoltaic power. The study now applies the chosen feature components as inputs into the optimized ANN with PSO to create the PV prediction model. The prediction strength of the proposed NCA-PSO-ANN model is compared with other variant hybrid models such as Principal Component Analysis (PCA), PSO and ANN (PCA-PSO-ANN), PSO-ANN, and PCA-ANN. The proposed NCA-PSO-ANN model constitutes a very reliable computational tool as it performed better in the selected performance indexes than the compared models.

Electrical energy, due to immediate consumption requirements, is a type of energy that needs to be planned, produced, and transmitted in a quick and efficient way. Short-term and long-term electricity forecasts are vital for energy providers, regulators, and consumers when planning and operating grids, estimating tariffs, and calculating energy supply and demand. This chapter explores the applications of artificial intelligence in short-term and long-term forecasting techniques including, but not limited to, statistical methods to deep learning algorithms. Various artificial intelligence algorithms and methods used in the applications in the spectrum of time-dependent electricity consumption to smart grid planning and operations are investigated extensively. Moreover, the properties of time series data, fundamental concepts of machine learning, and techniques of hyperparameter tuning are described in detail.