Investigation of electrical characteristics of residential light bulbs in load modelling studies with novel similarity score method
2020; Institution of Engineering and Technology; Volume: 14; Issue: 23 Linguagem: Inglês
10.1049/iet-gtd.2020.0674
ISSN1751-8695
Autores Tópico(s)Energy Load and Power Forecasting
ResumoIET Generation, Transmission & DistributionVolume 14, Issue 23 p. 5364-5371 Research ArticleFree Access Investigation of electrical characteristics of residential light bulbs in load modelling studies with novel similarity score method Rasim Doğan, Corresponding Author rsmdgn@gmail.com orcid.org/0000-0003-2122-9528 Electrical Engineering Department, Afyon Kocatepe University, Afyonkarahisar, TurkeySearch for more papers by this authorEmre Akarslan, Electrical Engineering Department, Afyon Kocatepe University, Afyonkarahisar, TurkeySearch for more papers by this author Rasim Doğan, Corresponding Author rsmdgn@gmail.com orcid.org/0000-0003-2122-9528 Electrical Engineering Department, Afyon Kocatepe University, Afyonkarahisar, TurkeySearch for more papers by this authorEmre Akarslan, Electrical Engineering Department, Afyon Kocatepe University, Afyonkarahisar, TurkeySearch for more papers by this author First published: 12 October 2020 https://doi.org/10.1049/iet-gtd.2020.0674Citations: 2AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onEmailFacebookTwitterLinked InRedditWechat Abstract In load modelling studies, the behaviour and electrical characteristic of a load are the main entries used in the identification of that load. This study is focused on the electrical characteristics of light bulbs (compact fluorescent light, fluorescent, incandescent, and light-emitting diode) with different characteristics that can be found in a house. While discovering these characteristics, current waveforms were used and electrical features were revealed. In addition, a novel method is proposed to identify individual loads by using these features. In this scope, a novel similarity score measure is introduced. In the presented study, similarity score measure is calculated from the total harmonic distortion, power, and harmonic current features. The experiments on randomly selected loads demonstrate that the proposed method successfully identified the loads even if they have similar characteristics or power consumption. 1 Introduction With the rapid development of technology, smart energy meter transformation has also accelerated [1]. The main objectives of this transformation can be listed as demand response [2], home energy management [3], demanding more information from the electric consumption of users, such as the health condition of customers [4]. To fulfil the objectives, loads must be monitored. This can be done either by placing sensors for each load or by measuring from outside the house. Since the sensor placement is an extra cost for utility and requires the permission of the customer, load monitoring with a single measurement has become more prominent. In 1992, Hart proposed a non-intrusive load appliance monitoring [5]. The main purpose was to identify appliances with the recorded power consumption (real and reactive power) at the utility side. A clustering algorithm, which is basically tracking the power consumption, was used as a method. Then, measurements were decomposed, and appliances and their quantities were estimated. Similar to Hart's study, the event window-based method also uses the clustering algorithm [6]. The method has benefitted from five signatures and these are edge, sequence, phase, time, and trend data. For example, the active time span of a light bulb is between sunset and sunrise. However, for the microwave, this will be breakfast, lunch, or dinnertime. So, this is one of the event windows for appliances. Another method uses self-configuring event detection (active power and transient power), which does not need parameter-tuning for any new environment [7]. The common point for the non-intrusive load monitoring (NILM) techniques using the clustering algorithm is to require continuous measurement of power and they are effective for high-power appliances (100 W and up). In addition to low-frequency power signatures (power consumption), a statistical approach such as a hidden Markov model (HMM) was also performed to estimate appliances from the aggregated signal [8, 9]. Factorial HMM [10] and additive factorial approximate maximum a-posteriori [11] are the derivatives of HMM. Although all these methods are successful in modelling, they require a large number of training data. Also, the number of training data increases incrementally for each new load addition. Another method uses a probabilistic approach for energy consumption of a group of electrical appliances [12]. The method requires long-time measurement and it is sensitive for a power interruption as in the clustering algorithm. Multi-label classification is also performed for NILM studies with extra electrical measurements (current, power factor, and reactive power) and its performance to identify lower power lightings has a large range starting at 59% [13]. Another feature used in NILM studies is the high-frequency (∼kHz) data, which is current and voltage waveforms [14–16]. Each appliance may have different current waveforms depending on the location and their electrical structure. Therefore, voltage and current trajectories have been used for the transfer learning [17]. However, the complexity of the method and training requirement is challenging parts. Instead of steady-state current waveform analysis, transient currents are the other option. These currents occur at the time of loads that are energised and disappear in a very short duration. Therefore, it is challenging to catch the waveform, and it supplies limited information for a load [18–20]. Thus, to record and analyse steady-state current and voltage waveforms are relatively easier than transients [21]. In addition, they contain more specific information about loads, such as harmonics [22]. Harmonics extracted by fast Fourier transform (FFT) of current waveforms have been used for a couple of loads and demonstrated that each load has a particular harmonic spectrum [23]. This unique spectrum was proposed to use as a feature. The amplitudes and phase angles of the current harmonics and their power consumption are considered for the training of the neural network [24, 25]. Results indicated that the reliability is susceptible to the training data and the method that is used. Self-organisation map method is used for the NILM and performed with electrical signatures of loads such as power factor, root mean square (rms) current, total harmonic distortion (THD), crest factor, and so on [26, 27]. This method suffers from the identification of similar loads that have the same power consumption, and also needs a very high amount of training data and user interface. The state-of-the-art does not have a complete solution for NILM studies because all methods have a trade-off. In addition, the identification of low-power loads has been ignored for most of the study. Considering all, the success rate of NILM studies by ignoring the electrical characteristics is very low. Therefore, this paper presents the investigation of electrical characteristics for different kinds of residential light bulbs as a low-power load and comparison of them with respect to electrical characteristics for the first time. Furthermore, the novel similarity score (SS) method is proposed. Therefore, power consumptions, THDs, fundamental rms currents, third, fifth, seventh, and ninth harmonic currents are considered for the SS method. For the verification, randomly selected load measurements have been successfully identified. The paper is organised as follows. Section 2 introduces the electrical characteristics and their calculations. Section 3 represents the test results and calculated quantities. Section 4 provides the identification method and results. Finally, Section 5 summarises the contribution. 2 Materials and method Light bulbs are manufactured by a wide range of materials, shapes, and light intensities. Fluorescents and incandescent light bulbs have been the most popular light sources for years. Nowadays, compact fluorescent light (CFL) and light-emitting diodes (LEDs) are getting popular in residential buildings as an energy saver light source. To understand the features of the low-power (under 100 W) lights; six CFLs, five incandescent, five LED, and one fluorescent light bulbs are investigated and they are listed in Table 1. Table 1. Residential light bulbs and their nominal powers Type Brand Nominal power, W CFL Osram 23 Fujika 13 Powerkey 20 Philips 20 Philips 8 Cimri 35 incandescent Osram 100 Osram 25 Philips 40 Tungsten 40 Vintage 40 LED Solarens 40 Glacial 30 Solarens 10 Bioplus 5 Osram 8.5 fluorescent Philips 2 ´ 36 Some loads were imported from the WHITED dataset [28] as a ‘flac’ file and their sampling rate is 44,100, and the rest was recorded as a ‘csv’ file with 250,000 sampling rate. A computer was used for calculations and it has Intel® Core™i5-6200U CPU@ 2.30 GHz processor and 8 Gb RAM and all calculations are performed by Matlab software. The first step is to obtain one period current waveform in comparison with zero-crossing voltage (see Fig. 1). This step has a critical role in the comparison of all data. Then, FFT is applied to the obtained current waveform for calculating harmonic components. The odd ones are only considered for the identification process since these are determinant and relatively larger than others. Also, any harmonic order above 9 does not contribute to the process. Therefore, third, fifth, seventh, and ninth harmonic components are considered. Fig. 1Open in figure viewerPowerPoint Measurement setup The second step is the calculation of THD with the below formula based on IEEE Std 519-1992 [29] ((1)) where THDIis the THD of current, I1is the fundamental current, and Ihis the harmonic current. The third step is the calculation of SS which is firstly proposed in this study. At this step, three filters are calculated; power consumption, THD, and harmonic distribution. Each filter generates a SS between all data located at the database and the measured data using (2)–(4). Then, the overall SS is calculated by (5). Thus, the measured load will be identified as the one that has the highest score by (6) ((2)) ((3)) ((4)) ((5)) ((6)) where Pdj and THDdj are the power and THD of the jth load in the database. Pm and THDmare the power and THD of measured load. Hmi is the ith harmonic of the measured load. Hdij is the ith harmonic of the jth load in the database. SSPj is the SS of power, SSTHDj is the SS of THD, and SSHPj is the SS of harmonic pattern for jth load in the database. 3 Measurements and harmonic analysis This section includes the current waveform of loads and their FFT analysis results. The measurement setup is also presented in Fig. 1. The data acquisition setup has one current and voltage probes. Each type of light is investigated in individual subsection. Also, voltage reduction measurements are presented in a separate subsection. 3.1 Compact fluorescent light The working principle of CFL is different than incandescent and LED light bulbs. Electric current circulates in a tube contains argon and mercury vapours and creates light. To regulate the electric current, the ballast is used. The design and operation of the ballast may result in different electrical characteristics. Current waveforms of six CFLs, five different brands, are presented in Fig. 2. Their FFT analysis results are given in Table 2. The fundamental (50 Hz) component is provided as rms current in ampere and third, fifth, seventh, and ninth harmonics are presented as the ratio of each to the fundamental component. Table 2. THD and harmonic distribution of current waveform of CFL lights Brand THD, % Harmonic Order 1st, A 3rd, % 5th, % 7th, % 9th, % Osram-CFL-23 W 102 0.11 72 41 24 25 Fujika-CFL-13 W 126 0.05 76 48 26 21 Powerkey-CFL-20 W 108 0.09 73 40 25 26 Philips-CFL-20 W 104 0.09 72 40 24 25 Philips-CFL-8 W 105 0.04 71 42 24 24 Cimri-CFL-35 W 106 0.18 70 38 26 25 Fig. 2Open in figure viewerPowerPoint Current waveforms of various CFLs Based on Fig. 2, all waveforms track a similar current path even though they have various brands and different power ratings. This reveals the expectation of similar harmonic distribution. Therefore, FFT analysis is performed for all, and results are presented in Table 2. This indicates two important results; similar THD and correlation of harmonic distribution pattern. First, the THDI of all CFL lights is above 100% and the largest THDI is 126%, the lowest one is 102%, and the average is 109%. This could be used as an important indicator to identify particular loads. Second, they have similar harmonic distribution patterns. It is presented in Fig. 3 and it is clearly seen that Fig. 3Open in figure viewerPowerPoint Harmonic distribution for CFL lights errors for all harmonics are low third harmonic is greater than all fifth harmonics is greater than seventh and ninth one seventh and ninth harmonics are similar. This harmonic distribution will be used in the identification process for all CFL light bulbs. 3.2 Incandescent Incandescent bulbs are one of the most common lights in the world and they consume large power to produce affordable light intensity. Light is generated by current driven through a thin filament. Besides producing light, they dissipate considerable heat. Incandescent light bulbs do not need a ballast or driver circuit since they consist of a resistance. The current waveforms of five different incandescent light bulbs are presented in Fig. 4 and their FFT analysis results are given in Table 3. Table 3. THD and harmonic distribution of current waveform of incandescent light bulbs Brand THD, % Harmonic order 1st, A 3rd, % 5th, % 7th, % 9th, % Osram-Inc-100 1 0.42 0 0 1 0 Osram-Inc-25 2 0.10 2 0 1 0 Philips-Inc-40 8 0.18 7 4 2 1 Tungsten-Inc-40 3 0.18 2 2 1 1 Vintage-Inc-40 1 0.18 0 0 1 1 Fig. 4Open in figure viewerPowerPoint Current waveforms of various incandescent light bulbs Both Fig. 4 and Table 3 demonstrate that all incandescent light bulbs have similar characteristics even they have different power ratings. The highest THDI is 8% and the lowest THDI is 1%. In addition, they have similar harmonic distribution patterns. It is presented in Fig. 5 and it is clearly seen that Fig. 5Open in figure viewerPowerPoint Harmonic distribution for incandescent lights errors for all harmonics are low all harmonics have very low amplitude. This harmonic distribution will be used in the identification process for all incandescent light bulbs. 3.3 Light-emitting diode LEDs consume low power and produce quite intense light. The light intensity is sensitive to voltage fluctuations. Therefore, LED lights should have a driver circuit to supply the same and pure DC voltage. Depending on the driver circuit, their current waveforms and electrical characteristics may differ. To show the difference, five LED lights are investigated in this study, and their current waveforms are shown in Fig. 6. Also, their FFT analysis results are presented in Table 4. Table 4. THD and harmonic distribution of current waveform of LED lights Brand THD, % Harmonic order 1st, A 3rd, % 5th, % 7th, % 9th, % Solares-LED-40 8 0.22 4 4 5 5 Glacial-LED-30 23 0.13 21 3 6 6 Solares-LED-10 11 0.05 5 5 2 7 Bioplus-LED-5 50 0.06 15 14 7 3 Osram-LED-8.5 103 0.04 69 44 23 21 Fig. 6Open in figure viewerPowerPoint Current waveforms of various LED lights LED light bulbs do not have similar harmonic distribution. For example, the highest THDI is 103% and the lowest one is 8%. This difference could be explained by the variation of their drivers. As aforementioned before, the driver circuit of each light source has shown dissimilarity except the same brand LED lights such as Solarens brand. Although one of them has four times higher power consumption than the other, their harmonic distributions are similar. Also, the harmonic distribution of Osram brand LED light resembles the same brand CFL lights. For better understanding, errors for all harmonics are presented in Fig. 7. This much high error could be an advantage for the method. Once it is measured and saved all data into the database, the SS method will be able to identify any LED lights because of their harmonic distributions. Fig. 7Open in figure viewerPowerPoint Harmonic distribution for LED lights 3.4 Fluorescent While fluorescent has had a wide usage area, as a result of the developments in technology, it has been replaced by light bulbs with lower power consumption in the same light flux such as CFL and LED. Since fluorescents still in use, they are included in the article, and the current waveform is presented in Fig. 8. Fig. 8Open in figure viewerPowerPoint Current waveforms of a fluorescent After Fourier analysis, THD is calculated as 14.56% while the fundamental current is 0.7234 Arms. Third, fifth, seventh, and ninth harmonic currents are calculated as 14.1, 0.51, 1.92, and 1.12% (see Fig. 9) with respect to fundamental current, respectively. The error between measurements is pretty low and it means the electrical character of fluorescent bulbs is stable. In addition, Fourier analysis indicates that the third harmonic is the distinctive feature for the fluorescent light. Fig. 9Open in figure viewerPowerPoint Harmonic distribution for fluorescent 3.5 Under voltage reduction The power system is not stable most of the time. For Turkey, the rms secondary distribution voltage is 220 V. However, this voltage may be less or higher than the desired level. In addition, regulations do not allow the end-user voltage reduction under 5% (209 V) for normal operation and 10% (198 V) for an emergency. Therefore, reduced voltage measurements are also performed for available loads to cover these situations and results are presented in Tables 5 and 6. Table 5. Five per cent voltage reduction test results Brand THD, % Harmonic order 1st, A 3rd, % 5th, % 7th, % 9th, % Philips-CFL-20 94 0.10 70 36 27 23 Osram-CFL-23 94 0.11 69 38 26 22 Fujika-CFL-13 122 0.05 77 49 31 28 Powerkey-CFL-20 105 0.09 72 44 30 25 Bioplus-LED-5 82 0.05 21 13 3 3 Osram-LED-8.5 105 0.05 76 46 28 23 Table 6. Ten per cent voltage reduction test results Brand THD, % Harmonic order 1st, A 3rd, % 5th, % 7th, % 9th, % Philips-CFL-20 97 0.09 72 38 27 25 Osram-CFL-23 92 0.11 68 37 26 22 Fujika-CFL-13 121 0.05 77 50 31 27 Powerkey-CFL-20 98 0.1 70 40 28 23 Bioplus-LED-5 91 0.05 21 13 4 2 Osram-LED-8.5 113 0.04 78 51 32 25 4 Results and discussion After all analysis, a method is developed to identify any measurement. The method contains the following six steps: Step 1: measure the voltage and current waveform of the load. Step 2: take only one period current waveform starts with zero-crossing voltage. Step 3: perform FFT analysis. Step 4: find the percentage similarity with power, THD, and harmonic distortion, respectively. Step 5: find the SS. Step 6: identify the load. If all these steps are applied to any measurement, the measured load would be identified accurately. As an example, a randomly selected load (which is Osram-CFL-23 W) was measured and results were evaluated by the method. Similarity results are presented in Table 7 and it contains the highest five SSs for each filter. As a result, Osram-CFL-23 W has the highest SS, which is 94%. In other words, the randomly selected load should be Osram-CFL-23 W with the probability rate of 94%. The probability rate is not 100% because the measurement has been performed at a different time and this creates a small error at measurements. Note that the selected load consumes 23 W power. Therefore, this much error on measurement reduces the SS. Table 7. Osram-CFL-23 W identification Similarity, % Power THD Harmonic Dist. Score Osram-CFL-23 W 92 96 95 94 Philips-CFL-20 W 76 94 95 88 Powerkey-CFL-20 W 76 91 95 87 Philips-CFL-8 W 33 93 93 73 Osram-LED-8.5 W 33 95 89 72 Glacial-LED-30 W 91 23 22 45 Osram-Inc.-25 W 83 3 2 29 Another example is used for a randomly selected incandescent light bulb (which is Vintage-Inc-40 W). Similarity results are presented in Table 8. As a conclusion, the Vintage-Inc-40 W obtains the highest SS, which is 100%. Table 8. Vintage-Inc-40 W identification Similarity, % Power THD Harmonic Dist. Score Vintage-Inc-40 W 99 100 100 100 Osram-Inc-100 W 43 100 100 81 Tungsten-Inc-40 W 99 33 100 77 Osram-Inc-25 W 54 50 100 68 Philips-Inc-40 W 99 13 50 54 Solarens-CFL-40 W 99 13 20 44 Glacial-CFL-30 W 72 4 16 31 Incandescent, fluorescent, and CFL light bulbs have a similar harmonic distortion in their own group. Even they have similar harmonic patterns and consume the same power, all light bulbs have been perfectly identified with the identification process. In addition, the identification process is performed on LED light bulbs and results are excellent even they have different values from each other for each filter. The experimental results illustrate the effectiveness of the proposed approach. As mentioned before the approach uses three different similarity measures to calculate the SS. When Table 7 is analysed carefully, the importance of this approach can be seen obviously. If only similarity measure of harmonic distribution is used for identification, Osram-CFL-23 W, Philips-CFL-20 W, and Powerkey-CFL-20 W will be determined as the same loads. Similarly, if the only similarity of power is considered, Philips-CFL-20 W and Powerkey-CFL-20 W will be evaluated in the same class while Philips-CFL-8 W and Osram-LED-8.5 W loads are assigned to another class. Therefore, to discriminate Philips-CFL-20 W and Powerkey-CFL-20 W or Philips-CFL-8 W and Osram-LED-8.5 W are not possible by using one of these similarity measures. However, the proposed SS can discriminate these loads successfully. The same situation can be seen even if Table 8 is analysed. If only power similarity is considered, Vintage-Inc-40 W, Tungsten-Inc-40 W, Philips-Inc-40 W, and Solarens-CFL-40 W are evaluated as the same class while Vintage-Inc-40 W and Osram-Inc-100 W or Philips-Inc-40 W and Solarens-CFL-40 W are considered as the same if only THD similarity is used. In the case of harmonic distribution similarity usage, Vintage-Inc-40 W, Osram-Inc-100 W, Tungsten-Inc-40 W, and Osram-Inc-25 W will be in the same class. It is obvious from these results that one similarity measure cannot discriminate the loads used in this study while the proposed approach can distinguish. The experimental results show that the proposed method can separate loads of the same power and this is another superior aspect of the approach. In Table 7, Philips-CFL-20 W and Powerkey-CFL-20 W, in Table 8, Vintage-Inc-40 W, Tungsten-Inc-40 W, Philips-Inc-40 W, and Solarens-CFL-40 W are in the same power; however, their SSs are different. This means that the proposed approach can distinguish these loads. Furthermore, this approach utilises from SS and does not need a training stage. In load identification literature, to distinguish the small power loads is challenging. The power range of loads in this study changes from 8 to 100 W and the proposed approach identifies all of them under nominal voltage conditions. This is another advantage and contribution of the approach. Since the SS is used, the computational complexity of the method is low. In the case of voltage fluctuations, the performance of the method is also investigated. In this scope, Osram-LED-8.5 and Bioplus-LED-5 are considered for the LED type light source. Results are given in Tables 5 and 6, respectively. SSs are applied to voltage reduction measurements and Bioplus-LED-5 can be precisely identified. However, Osram-LED-8.5 is identified as Philips-CFL-8 W. In addition, another mismatch is for the identification of Powerkey-CFL-20. It is identified as Philips-CFL-20. Although the brand of the CFL is not correctly determined, the type of light source and the power rating are correctly identified. To evaluate the overall performance of the method, accuracy, precision, recall, and F-score metrics are used. The accuracy is calculated by using (7) while the precision, recall, and F-score values are calculated by using (8)–(10), respectively [30] ((7)) ((8)) ((9)) ((10)) where NE is the number of experiments, NSI is the number of successful identifications, TP is the true positive (detected condition when the condition is present), TN is the true negative (not detected condition when the condition is absent), FP is the false positive (detected condition when the condition is absent), and FN is the false negative (not detected condition when the condition is present). The total (considering voltage fluctuations as well) performance of the proposed method is presented in Table 9. All metrics, Accuracy, F-score, precision, and recall, are over 97% and it means that the method provides successful results even under voltage fluctuations. Table 9. Total performance of the proposed method Percentage of accuracy, % Precision Recall F-score total performance 97.8 0.978 0.978 0.978 5 Conclusions Four different types of light sources, LED, CFL, fluorescent, and incandescent light bulbs, are investigated based on their current waveforms, and a novel identification method is presented for NILM studies. A new measure called a SS is introduced in this study. The proposed novel method uses SS, which is calculated from three similarity measures of THD, harmonic currents, and power. The results show that the approach can perfectly identify the loads with similar characteristics or even the same power rates with only one cycle measurement thanks to the three-stage similarity approach. It is illustrated that one similarity measure is not enough to distinguish loads used in the study and this shows the importance of the proposed approach. The novel approach does not need a training phase and can distinguish the loads with small power even under voltage fluctuations. Furthermore, the computational complexity of the method is low since SS measure does not contain complex calculations. Hence, the overall performance of the method is above 97%, which illustrates the success of the method. For future works, a real-time identification hardware, which has embedded software, will be designed. The aim of the hardware, also, is to identify any appliances (not limited to light sources) located at residential monitored by the hardware. 6 References 1Sun, Q., Li, H., Ma, Z. et al.: ‘A comprehensive review of smart energy meters in intelligent energy networks’, IEEE Internet Things J., 2016, 3, (4), pp. 464– 479CrossrefWeb of Science®Google Scholar 2Arun, S.L., Selvan, M.P.: ‘Intelligent residential energy management system for dynamic demand response in smart buildings’, IEEE Syst. J., 2018, 12, (2), pp. 1329– 1340CrossrefWeb of Science®Google Scholar 3Lin, Y.H., Tsai, M.S.: ‘An advanced home energy management system facilitated by nonintrusive load monitoring with automated multiobjective power scheduling’, IEEE Trans. Smart Grid, 2015, 6, (4), pp. 1839– 1851CrossrefWeb of Science®Google Scholar 4Alcala, J.M., Urena, J., Hernandez, A. et al.: ‘Sustainable homecare monitoring system by sensing electricity data’, IEEE Sens. J., 2017, 17, (23), pp. 7741– 7749CrossrefWeb of Science®Google Scholar 5Hart, G.W.: ‘Nonintrusive appliance load monitoring’, Proc. IEEE, 1992, 80, (12), pp. 1870– 1891CrossrefWeb of Science®Google Scholar 6Dong, M., Meira, P.C.M., Xu, W. et al.: ‘An event window based load monitoring technique for smart meters’, IEEE Trans. Smart Grid, 2012, 3, (2), pp. 787– 796CrossrefWeb of Science®Google Scholar 7Afzalan, M., Jazizadeh, F., Wang, J.: ‘Self-configuring event detection in electricity monitoring for human-building interaction’, Energy Build., 2019, 187, pp. 95– 109CrossrefWeb of Science®Google Scholar 8Guo, Z., Wang, Z.J., Kashani, A.: ‘Home appliance load modeling from aggregated smart meter data’, IEEE Trans. Power Syst., 2015, 30, (1), pp. 254– 262CrossrefWeb of Science®Google Scholar 9Kong, W., Dong, Z.Y., Ma, J. et al.: ‘An extensible approach for non-intrusive load disaggregation with smart meter data’, IEEE Trans. Smart Grid, 2018, 9, (4), pp. 3362– 3372CrossrefWeb of Science®Google Scholar 10Egarter, D., Bhuvana, V.P., Elmenreich, W.: ‘PALDi: online load disaggregation via particle filtering’, IEEE Trans. Instrum. Meas., 2015, 64, (2), pp. 467– 477CrossrefWeb of Science®Google Scholar 11Ji, T.Y., Liu, L., Wang, T.S. et al.: ‘Non-intrusive load monitoring using additive factorial approximate maximum a posteriori based on iterative fuzzy c -means’, IEEE Trans. Smart Grid, 2019, 10, (6), pp. 6667– 6677CrossrefWeb of Science®Google Scholar 12Liu, Y., Geng, G., Gao, S. et al.: ‘Non-intrusive energy use monitoring for a group of electrical appliances’, IEEE Trans. Smart Grid, 2018, 9, (4), pp. 3801– 3810CrossrefWeb of Science®Google Scholar 13Buddhahai, B., Wongseree, W., Rakkwamsuk, P.: ‘A non-intrusive load monitoring system using multi-label classification approach’, Sustain. Cities Soc., 2018, 39, pp. 621– 630CrossrefWeb of Science®Google Scholar 14Laughman, C., Lee, K., Cox, R. et al.: ‘Power signature analysis’, IEEE Power Energy Mag.., 2003, 1, (2), pp. 56– 63CrossrefGoogle Scholar 15Ito, M., Uda, R., Ichimura, S. et al.: ‘A method of appliance detection based on features of power waveform’. Proc. Int. Symp. on Applications and the Internet, Tokyo, Japan, 2004, pp. 291– 294Google Scholar 16Li, J., Yang, H.: ‘The investigation of residential load identification technologies’. Asia-Pacific Power and Energy Engineering Conf., APPEEC, Shanghai, China, 2012 Google Scholar 17Liu, Y., Wang, X., You, W.: ‘Non-intrusive load monitoring by voltage-current trajectory enabled transfer learning’, IEEE Trans. Smart Grid, 2019, 10, pp. 5609– 5619, doi: 10.1109/TSG.2018.2888581CrossrefWeb of Science®Google Scholar 18Chang, H.-H., Lian, K.-L., Su, Y.-C. et al.: ‘Power-spectrum-based wavelet transform for nonintrusive demand monitoring and load identification’, IEEE Trans. Ind. Appl., 2014, 50, (3), pp. 2081– 2089CrossrefWeb of Science®Google Scholar 19Wu, X., Han, X., Liang, K.X.: ‘Event-based non-intrusive load identification algorithm for residential loads combined with underdetermined decomposition and characteristic filtering’, IET Gener. Transm. Distrib., 2019, 13, (1), pp. 99– 107Wiley Online LibraryWeb of Science®Google Scholar 20Wu, X., Han, X., Liu, L. et al.: ‘A load identification algorithm of frequency domain filtering under current underdetermined separation’, IEEE Access, 2018, 6, pp. 37094– 37107CrossrefWeb of Science®Google Scholar 21Aladesanmi, E.J., Folly, K.A.: ‘Overview of non-intrusive load monitoring and identification techniques’, IFAC-PapersOnLine, 2015, 48, (30), pp. 415– 420CrossrefGoogle Scholar 22Dong, M., Meira, P.C.M., Xu, W. et al.: ‘Non-intrusive signature extraction for major residential loads’, IEEE Trans. Smart Grid, 2013, 4, (3), pp. 1421– 1430CrossrefWeb of Science®Google Scholar 23Akbar, M., Khan, Z.A.: ‘Modified nonintrusive appliance load monitoring for nonlinear devices’. 2007 IEEE Int. Multitopic Conf., Lahore, Pakistan, 2007, pp. 1– 5Google Scholar 24Yoshimoto, B., Nakano, K., Amano, Y. et al.: ‘Non-intrusive appliances load monitoring system using neural networks’. ACEEE Buildings Conf. Proceeding Book, Washington, DC, USA, 2000, pp. 183– 194Google Scholar 25Srinivasan, D., Ng, W.S., Liew, A.C.: ‘Neural-network-based signature recognition for harmonic source identification’, IEEE Trans. Power Deliv., 2006, 21, (1), pp. 398– 405CrossrefWeb of Science®Google Scholar 26Du, L., He, D., Yang, Y. et al.: ‘Self-organizing classification and identification of miscellaneous electric loads’. IEEE Power and Energy Society General Meeting, San Diego, CA, USA, 2012 Google Scholar 27Du, L., Restrepo, J.A., Yang, Y. et al.: ‘Nonintrusive, self-organizing, and probabilistic classification and identification of plugged-in electric loads’, IEEE Trans. Smart Grid, 2013, 4, (3), pp. 1371– 1380CrossrefWeb of Science®Google Scholar 28Matthias Kahl, H.-A.J., Ul Haq, A., Kriechbaumer, T.: ‘WHITED-A worldwide household and industry transient energy data set’. Proc. 3rd Int. Workshop Non Intrusive Load Monitoring (NILM), Vancouver, Canada, 2016, pp. 1– 4Google Scholar 29519-1992 IEEE Recommended Practices and Requirements for Harmonic Control in Electrical Power Systems Google Scholar 30Andrean, V., Zhao, X.H., Teshome, D.F. et al.: ‘A hybrid method of cascade-filtering and committee decision mechanism for non-intrusive load monitoring’, IEEE Access, 2018, 6, pp. 41212– 41223CrossrefWeb of Science®Google Scholar Citing Literature Volume14, Issue23December 2020Pages 5364-5371 FiguresReferencesRelatedInformation
Referência(s)