Internet of energy-based demand response management scheme for smart homes and PHEVs using SVM

Highlights

• Unified solution to manage the demand response of SHs and PHEVs in IoE network.

• A load balancing algorithm to curtail the load in SHs identified using SVM classifier.

• A binary-class SVM classifier to identify PHEVs whose charging rates can be altered.

Abstract

The usage of information and communication technology (ICT) in the power sector has led to the emergence of smart grid (SG). The connected loads in SG are able to communicate their consumption data to the grid using ICT and thus forming a large Internet of Energy (IoE) network. However, various issues such as–increasing demand–supply gap, grid instability, and deteriorating quality of service persist in this network which degrade its performance. These issues can be handled in an efficient way by managing the demand response (DR) of different types of loads. For this purpose, cloud computing can be leveraged to gather the data generated in IoE network and perform analytics to manage DR. Working in this direction, a novel scheme to handle the DR of smart homes (SHs) and plug-in hybrid electric vehicles (PHEVs) is presented in this paper. The proposed scheme is based on analyzing the demand of these users at the cloud server for flattening the overall load profile of grid. This scheme is divided into two hierarchical stages which work as follows. In the first stage, the residential and PHEV users are identified whose demands can be regulated. This task is achieved with the help of a binary-class support vector machine (SVM) which uses Gaussian kernel function to classify these users. In the next stage, the load in SHs is curtailed on the basis of a pre-defined rule-base after analyzing the consumption data of various devices; whereas PHEVs are managed by controlling their charging rates. The efficacy of proposed scheme has been tested on PJM benchmark data and Open Energy Information dataset. The simulation results prove that the proposed scheme is effective in maintaining the overall load profile of SG by managing the DR of SHs and PHEV users.

Introduction

The future of power industry heavily relies on the usage of modern electric grids integrated with information and communication technology (ICT). These types of grids are commonly known as smart grids (SGs) [1]. The main objective of SG is to provide better quality of service to the consumers by reducing the demand–supply gap. One of the techniques to accomplish this objective is by managing the consumers’ demand at all the times; such that the load is reduced during peak time (i.e., peak shaving) and increased during off-peak times (i.e., valley filling) [2]. This task can be fulfilled by efficiently handling the demand response (DR) of various types of sectors such as-industrial, commercial, and residential. Due to technological advancements, the devices placed inside the homes, buildings and offices are equipped with ICT and other sensors which constantly monitor and send the usage data of these devices to the local controller. This local controller can then forward the energy usage data of various devices to the utility server in order for it to analyze and process. As these devices are huge in number and are able to send the energy data at very small time intervals, the complete device network can be viewed as one large Internet of Energy (IoE) network in the SG. Thus, in this IoE network, the devices send their energy usage data to the central server which is analyzed and DR of the connected loads is managed in an effective way so that the demand and supply gap can be reduced.

As far as residential loads are concerned, researchers have proposed different methodologies to manage their demands. For example, Zhou et al. [3] used auction-based strategy to manage the demand at grid. The authors invited bids from the end users to shed their loads when supply was less and accepted bids which maximized the grid benefits with minimum additional costs. Working in the similar direction, Safdarian et al. [4] proposed a decentralized approach to manage the load profile of homes such that the overall load stability on the grid can be improved. The authors used mixed integer linear programming to solve the load optimization problem in SHs. The drawbacks of this approach are that it may take time to converge to an optimal solution and consumers’ load were modified every time when the grid’s load profile was updated. Both of these conditions can hamper the smooth working of DR management process and may not be possible in a realistic scenario. In an another scheme, Costanzo et al. [5] presented a load management scheme in smart buildings to control the operations of appliances so as to minimize its overall consumption cost. In addition to these approaches, the home energy management systems have also been extensively employed by the researchers to manage the load of residential users so as to decrease their overall cost  [[6], [7]]. For example, in [6], the authors used Markov decision process to schedule the appliances in homes to decrease the power balancing cost in their neighborhood. In [7], the authors used greedy approach to manage the DR in homes to decrease the price of electricity. However, the simultaneous working of individual energy management systems in various SHs in an uncoordinated manner may result in a new rebound peak. Moreover, none of these schemes take the advantage of historical data gathered from smart devices in SHs to achieve load stability.

The industrial and commercial loads are not managed in the proposed scheme because any modification made in their electricity demand can severely affect their business and routine functioning. Apart from these loads, plug-in hybrid electric vehicles (PHEVs) also have a huge impact on the electricity market because of their increased popularity in the transportation sector. According to the Global EV Outlook, there will be around 20 million electric vehicles (EVs) on road by the year 2020 [8]. These EVs can be leveraged to manage the overall load on SG by regulating their charging requirements. Many studies exist in the literature which considered the load demand of PHEVs to manage their DR [[9], [10], [11]]. Authors in [9] proposed a mechanism for handling the charging rates of PHEVs’ batteries according to the available price and user’s willingness to pay. Similarly, a price control mechanism for charging infrastructure to manage the charging requests of PHEVs based on an auto-regression model was provided in [10]. In [11], authors proposed a peak load optimization strategy for PHEVs based on real-time prices to ease off the burden on grids during peak hours. However, all these studies focused on increasing the price during peak hours; but in reality, increased price may not always have a significant effect on the load demand of PHEV owners. It is because the PHEV owners might be willing to pay extra to get the charging facility. Hence, a novel scheme to cater the charging requirements of PHEVs without putting additional burden on the grid is required.

Moreover, the DR of residential loads and PHEVs can be handled more effectively with the help of data analytics, i.e., by analyzing their consumption data [12]. Many analytical algorithms such as-regression, classification, clustering algorithms, etc. have already been used in the power sector to mitigate various issues related to SG. For example, authors in [[13], [14], [15], [16]] have used a classification algorithm namely support vector machine (SVM) for demand forecasting, theft detection, and understanding patterns in consumer’s demand, respectively. Keeping this in view, a novel data analytical scheme has been presented in this paper to manage the DR of residential sector and PHEVs. However, the data gathered from these loads in the IoE network is continuous in nature and require abundant speed and storage space to transmit and analyze this data. Moreover, the data gathered in IoE network comes from heterogeneous devices, thus, it is essential to store and process this data in a manner that it becomes easy to analyze. For this purpose, the cloud computing paradigm can be leveraged as a platform to provide storage and processing capabilities. The advantage of using cloud computing paradigm is that it is easily scalable and accessible, have high throughput, and low cost of deployment & maintenance [[17], [18]]. Many authors have utilized cloud platform in order to gather and analyze this type of streaming heterogeneous data. For example, the authors in [19] proposed an elastic resource management scheme for streaming data analytics flows. The authors used advanced control and optimization techniques to design a control system for data input, applying analytics and storage layers of the proposed framework. In addition to it, many researchers have used parallel processing of file systems and other optimization-based techniques such as-particle swarm optimization for remotely gathering the temporal data from various heterogeneous devices [[20], [21]]. The similar techniques can be applied in this IoE network to gather and store the heterogeneous data at the cloud. However, as the main focus of the paper is to present the DR management scheme, therefore, the data collection process is not discussed in detail in this paper. The comparative analysis of the related works with proposed scheme is summarized in Table 1.

This paper is an extension of our earlier work [2] and provides additional functionalities in terms of managing the residential load in off-peak hours and handling the instantaneous load changes. In addition to the residential loads, the present work also handles the charging requirements of PHEVs by regulating their charging rates in such a way that the load profile of SG remains stable in peak and off-peak hours. The proposed scheme is different from existing schemes in literature in the following way. It uses SVM classifier to identify the residential and PHEV users whose load requirements can be regulated in order to fill the demand–supply gap. Once such users are identified, the residential DR is managed by analyzing their consumption data and load in SHs is curtailed according to a pre-defined rule-base. For managing the DR of PHEVs, SVM classifies them into slow and fast chargeable PHEVs whose charging rates are regulated during peak and off-peak hours respectively. The reason for selecting SVM for classification of users in the proposed scheme is because of its ability to perform better as compared to other classifiers in terms of accuracy and classification time, when data is non-linearly distributed [[13], [14], [15], [16], [22]].

The major contributions of the proposed scheme are summarized as follows.

• It provides a unified solution to manage the DR of residential loads and PHEVs in an IoE network by analyzing their demand requirements.

• A load balancing algorithm has been designed to curtail the load of residential users (identified using SVM classifier) on the basis of a pre-defined rule-base by analyzing their consumption data.

• A binary-class SVM classifier has been used to identify the PHEVs whose charging rates can be altered in order to flatten the load on SG.

The rest of the paper is organized as follows. Section 2 discusses the working of the proposed scheme. Section 3 elaborates on the residential load management technique while the PHEVs load handling technique is described in Section 4. Section 5 highlights the simulation results and the paper is finally concluded in Section 6.