Demand-Side Energy Management (EM) Techniques

In document Privacy-preserving energy management techniques and delay-sensitive transmission strategies for smart grids (Page 48-50)

1.2 Dissertation Outline and Research Contributions

2.2.6 Demand-Side Energy Management (EM) Techniques

In the aforementioned privacy-preserving techniques, privacy is provided by tampering the SM readings before being reported to the UP. As opposed to these techniques, demand-side EM is an emerging technique that can provide privacy to the consumer without tampering the SM readings. Demand-side EM techniques utilize storage units, such as RBs, and alterna- tive energy sources, such as a solar panel, to partially mask the energy usage patterns of the consumers against the UP. Moreover, since the SM readings are reported to the UP without tampering, these techniques maintain unaltered the operational utility of the SM readings.

Various EM algorithms have been proposed in the literature to provide privacy to users. In this regard, [85] proposes the best effort algorithm, which intends to hide the load signatures of the consumer from the UP with the utilization of RB. The proposed algorithm charges and discharges the RB in order to maintain a constant SM load level so that appliance usage events cannot be detected. The authors consider three different privacy metrics, namely, relative en- tropy, cluster classification and correlation/regression analysis, to measure the privacy provided by the proposed algorithm. In [86], a power mixing algorithm is proposed to protect energy consumption events of selected appliances with the utilization of RB. The authors consider the privacy metrics mentioned above and evaluate the performance of the proposed algorithm by using the SM data collected from individual home appliances. The authors indicate that some major factors, such as battery capacity and power, can have an effect on the performance of the proposed algorithm.

In [87], a simple RB system is studied. The authors consider a discrete-time system model with binary input-output loads and battery states, and propose stochastic battery policies to pro- vide privacy to the users. Mutual information between the input and output loads is considered as a measure of privacy. The authors compute the mutual information applying a trellis algo- rithm on the finite state model (FSM). They consider two types of stochastic policies, namely, battery-conditioned policies and battery/output-conditioned policies, and indicate that these policies can leak 26% less information than the algorithm proposed in [85].

In [88], the authors propose a novel technique for hiding sensitive power consumption signatures of the appliances in the total power consumption load of a household. The proposed method modifies the power consumption of the household through the utilization of RB con- nected to the household’s power supply, with the goal of providing privacy assurances in terms of differential privacy. The authors consider capacity and throughput constraints of batteries

2.2. Privacy-Preserving Techniques

in realistic scenarios, and propose an integrated method of noise cascading that maintains the differential privacy.

The authors in [89] propose a non-intrusive load leveling (NILL) algorithm to protect privacy of the user against the potential privacy invasion that can stem from NALM techniques. The proposed algorithm is used to flatten the consumption of the user to a constant target load, with the goal of removing appliances’ features. NILL uses RBs to flatten the power consumed by appliances. When an appliance turns to ON state, the exerted load exceeds the target load; and thus, NILL discharges the battery for partially satisfying the exerted load and maintaining the target load. Similarly, when an appliance turns to OFF state, the exerted load falls below the target load; and hence, NILM charges the battery with the energy drawn from the UP and maintains the target load. The proposed NILL system comprises an RB along with a control system that charges or discharges the RB based on the present load and battery state.

In [90], the authors propose three techniques, namely, fuzzing, targeted entropy maximiza- tion and targeted fuzzing. The proposed techniques intend to mask individual load changes with the utilization of RB. These techniques have different ways of choosing load offsets. The first proposed technique is fuzzing. This technique changes the observed load to a desired observed signal, which is chosen randomly over an interval by using a uniform distribution. At first glance, uniform distribution would seem to create the greatest obfuscation for an actual signal change. However, since the sampling interval is built around the actual load change, there are cases where this technique can choose an output signal value that has only one possible under- lying actual event. This would lead to the fact that there is no obfuscation at all. The second proposed technique is targeted entropy maximization. This technique chooses the desired load level that maximizes the entropy of possible individual events. To do so, the proposed tech- nique uses the information about the individual loads that contribute to the aggregate signal, and picks up an offset value to minimize the ability of the third parties to deduce any information about the individual appliances in the aggregate load. This technique assumes that the observer is unaware of the masking technique, the battery capacity and the charging/discharging rate. The authors indicate that this technique might fail in providing privacy if this information is available at the observer side, in which case the observer could decode the observed signal to reveal the original signal. The third technique is targeted fuzzing. This technique builds a prob- ability distribution for an observed event taking into account the fact that how this event can be interpreted by an observer. The distribution has bias towards samples that larger numbers of possible actual events can explain. This technique randomly samples an observed change from this distribution, while eliminating any samples that only one actual event can explain. The authors mention that the targeted fuzzing technique prevents the deficiencies of the previous techniques against potential attacks.

In [91], the authors propose battery-based load hiding methods to hide appliance loads. They first evaluate the performance of two well-known battery control algorithms, namely, best effort [85] and NILL [89], against the attack of an intruder, and reveal privacy vulnerabilities of these algorithms. Then, they propose a stepping-based algorithms based on maximizing the error between the input and output loads under the RB capacity and charging/discharging rate constraints. They use the mutual information as the privacy measure, and compare the performances of the proposed stepping algorithms against the best effort and NILL algorithms. Using a real energy consumption data, they show that their methods outperform the best effort and NILL algorithms in general.

The authors in [92] propose a stochastic control method that jointly decorrelates the SM readings from the user’s actual usage and reduces the energy cost of the user with the utilization of RB. The proposed method is founded based on a stochastic DP. The authors indicate that their method reduces the correlation between the SM readings and the user’s real consumption while maximizing the energy savings of the users. The cost savings are achieved by charg- ing the battery in the low-price zone and satisfy the energy demand from the stored battery energy in the high-price zone. According to their experiments, the authors indicate that the proposed technique achieves higher privacy and cost savings in the presence of low-frequency components in the load profile of the user.

In [93], the authors study a DP framework that jointly provides SM data privacy and reduces the energy cost of the user. Assuming that the energy demands and prices are known causally, they reformulate the original problem so that it can be solved by using only the current demand and price information. Then, they propose a low complexity online algorithm based on the Lyapunov optimization technique. The proposed algorithm is parametrized by a positive value, which enables to quantify the impact of the battery capacity on its performance. Using a real energy demand data, the authors demonstrate that their algorithm can provide privacy to the user in a cost-effective manner.

In document Privacy-preserving energy management techniques and delay-sensitive transmission strategies for smart grids (Page 48-50)