Thesis on the topic A Residential Load Scheduling Approach Based on Load Behavior Analysis
A Residential Load Scheduling Approach Based on Load Behavior Analysis Abstract— Residential load scheduling is of great significance for the building energy saving. But it is difficult for current methods to discover and consider the relationship between the various comfort requirements and uncertain load demand of customers. In this paper, a new method for the residential load scheduling based on load behavior analysis is proposed to reduce the electrical cost while considering the human behavior. Firstly, the multi-dimensional feature-based load profile identification is applied to identify the operating status of the appliances from the historical load data. Then, the environment record and the state models of appliances are used to estimate the comfort preferences and load demand of customers. The personalized optimal strategy of the household appliance operation is given to minimize the daily cost, according to the dynamic electrical price and the real time weather condition. Three simulations are designed to demonstrate how to identify the operation status of appliances, how to obtain the load behavior, and the performance of proposed method on cost saving. Index Terms--residential load scheduling, comfort requirements, load identification, load behavior analysis, dynamic electrical price I. INTRODUCTION Improving the efficiency in energy consumption is the most effective and the least controversial way to resolve or alleviate the worldwide energy crisis. Buildings contribute a significant part in the energy consumption. For example, there are more than 79 million residential buildings and 5 million commercial buildings in the United States of America, all of which consume about 40% of the energy and 70% of the electricity in the country [1]. In China, the buildings consume about 27%~29% of the final electricity [2]. As a large part of the energy consumption in buildings, the residential buildings are mainly focused on in this paper. Demand response (DR) is a potential solution for residential building energy saving, which is defined as the changes in electric usage of end-use customers from their normal consumption patterns in response to changes in the price of electricity [3]. In this process, participants shade or shift the electricity consumption during *Resrach supported by National Natural Science Foundation of China (61174146, 61221063, 61203174, 61304212, U1301254, 91118005), Doctoral Fund of Ministry of Education of China (20110201120010) and the Fundamental Research Funds for the Central University) Siyun Chen, Xiaohong Guan, Ting Liu, Yulin Che and Yuqi Liu are with the SKLMS Lab and MOE KLINNS Lab of Xi’an Jiaotong University, Xi’an, 710049, China (e-mail:
[email protected],
[email protected]; Ting Liu is the corresponding author: +8618691838686, e-mail:
[email protected]). Feng Gao is with the SKLMS Lab of Xi’an Jiaotong University, Xi’an, 710049, China (e-mail:
[email protected]). high-price periods and receive benefits from the low-price periods. Meanwhile, such a good schedule is benefit for alleviating a number of challenges faced by the power grid, including the peak load demand and generation supply shortage [4]. The residential load scheduling on demand side is not as complicated as the joint schedule of both the demand and the supply of power systems. However, the load scheduling for residential building is nontrivial due to the following difficulties. First, the dynamics of the various appliances in residential buildings usually cannot be described by precise models but only approximate models. And the power consumptions of the appliances have certain randomness. Second, the human comfort requirements and the load demand in buildings are all subject to random fluctuations, due to the changes of the environment conditions and human behavior. Third, the scheduling problem contains multiple stages. Meanwhile, with considering the randomness of the human behavior, the integer variables and the random variables will make it difficult to solve this scheduling problem. In this paper, a novel load scheduling approach is proposed for residential buildings exploiting the load behavior analysis. We make two major contributions. First, we improve the nonintrusive load monitoring (NILM) [5] method based on multi-dimensional load signatures to identify the operation status and electricity consumption of the appliances from the historical load data of each customer. And then combined with the historical weather data, the operation information of appliances are analyzed to find the distribution of operation and to obtain the most likely operation periods as the load demand periods. Second, the models of the appliances are established to estimate the work point of the appliances in common use as the comfort requirements, according to the operation information. In this novel framework of residential load scheduling with considering the randomness of human behavior, the personalized load control strategies are generated for each customer to minimize the electricity cost, according to the personalized load behavior, the current weather condition and dynamic electrical price. Moreover, due to the two major randomness factors including the load demand periods and comfort requirements are analyzed in advance, the solution of the scheduling problem is greatly simplified. In this paper, the air conditions and water heater are selected to simulate the experiments to demonstrate the proposed method due to the large proportion of household electricity consumption [6]. The rest of this paper is organized as follows. In section II, the related works are briefly reviewed. In section III, the detailed methods of load behavior analysis is presented. In A Residential Load Scheduling Approach Based on Load Behavior Analysis * Siyun Chen, Student Member, IEEE , Feng Gao, Member, IEEE , Xiaohong Guan, Fellow, IEEE , Ting Liu, Member, IEEE , Yulin Che, and Yuqi Liu 2014 IEEE International Conference on Automation Science and Engineering (CASE) Taipei, Taiwan, August 18-22, 2014 978-1-4799-5283-0/14/$31.00 ©2014 IEEE 954 section IV, the load scheduling problem is developed. In section V, the experiments demonstrate that about 20%~30% electricity cost could be saved. A brief conclusion is presented in section VI. II. LITERATURE REVIEW There are many researches on the load scheduling for energy saving in buildings. James E. Braun described the development of intelligent building system [7]. Ciara O’Dwyer [4] and D. P. Chassin [8] illustrate the potential of load control in residential sector on energy efficiency and the benefit for the operation of the power system. S. Tiptipakorn gave a rule based control strategy of appliances based on consumer predefined parameters such as temperature and electricity price [9]. A-H. Mohsenian-Rad proposed an incentive-based energy consumption scheduling approach to achieve a trade-off between minimizing the electricity cost and minimizing the waiting time for using the appliances [10]. However, the methods above have no consider the comfort level. S.Z.Althaher’s method considered the comfort level and assigned a random degree of flexibility and the freedom to customers [11]. Zhijin Cheng proposed an integrated control of blind and lights for energy saving with considering the change of environment conditions [12]. Biao Sun have modeled the integrated control of multiple devices, including HVACs, lights, shading blinds, and natural ventilation, and proposed a method to deal with the computational difficulty [13]. Xiaohong Guan and Zhanbo Xu mainly focus on the commercial buildings [14] and manufacturing industry [15], and they modeled the joint schedule of both the demand and multiple power generations. Their works on load schedule of the demand side considered the comfort requirements and state models of rooms. The load demand periods are necessary for load scheduling, while the existing methods suppose that the load demand periods are whole day, or constant. Most studies with considering the comfort requirements are also based on the similar assumption, without considering the uncertainty of human behavior and how to obtain the load behavior, especially for the residential customers,. III. LOAD BEHAVIOR ANALYSIS A. Load Identification The goal of NILM is to utilize the aggregate load profile to identify the operation status of appliances and estimate their power consumption. Generally, the operation of the appliances will produce the unique load signatures, such as the voltage and current waveforms (including harmonics) or processed reproductions of these signals such as the real power and reactive power. Different with the typical NILM [5], we use the multi-dimensional load signatures which are conveniently acquired by Smart Meter to improve the accuracy of the load identification. In our approach, the major improvements are the events detection with harmonics and the clustering based on multi-dimensional load signatures. 1) Events Detection with Harmonics The changes of the operation states will cause the changes of the load signatures, which can be considered as the events. The event detection aims to find the sufficient changes of these signatures while ignoring minor changes, as the first step to identify the appliances. Load signatures have different responsiveness to the events. Previous methods [5], [16] usually use the active power for the events detection with a certain threshold. However, for some appliances, the fluctuations of the active power produced in the operation process will be incorrectly detected as the events. For example, as shown in Fig.2, the fundamental current as the representation of the real power fluctuates largely. As a result, all the significant spikes in the power draw will be detected as the events. While the 3rd harmonic current is more stable and reflects the events well. So we combine the fundamental current and the 3rd harmonic current as the load signatures for events detection, and the arrows in Fig.2 show the exact events. It can be seen that detection accuracy is improved. 0 500 1000 1500 2000 2500 3000 3500 0 0.1 0.2 0.3 0.4 0.5 Time (s) C u rr en t (A ) Figure 1. The aggregate current of computer and lamp.(Blue line: fundamental current; Red line: the 3rd harmonic current; The arrows: the events) As a result, the input of our events detection method is a series of power data tuples, including {time stamp, fundamental current value and 3 rd harmonic current value}. The