Abstract
Accurate classification of power load is the premise of demand-side response capability evaluation, which is of great significance for demand-side management. Therefore, this study proposes a selection method of joint characteristics of power load in time domain and frequency domain based on whale optimization algorithm. The electrical measurement data of six typical electrical equipment are selected to form the original power load identification dataset. Firstly, the time domain features and frequency domain features of the power load data are extracted, and then the joint characteristics of the power load are screened by Whale optimization algorithm (WOA). Finally, the selected feature information is used as the input to verify the performance of WOA on the joint characteristics of power load under back propagation neural network (BP), extreme learning machine (ELM), support vector machine (SVM), k-nearest neighbor (KNN), decision tree (DT), and naive Bayesian (NB) classifiers. The results show that WOA can effectively screen out the 15 most helpful feature attributes for power load identification, which can not only improve the accuracy of power load identification but also effectively reduce the time cost of the algorithm, which is of reference value for further demand-side response strategy. At the same time, it is of great significance for intelligent dispatching of power system and improving the economic operation of the power system.
1. Introduction
With the vigorous development of energy Internet [1, 2], the development of industrial Internet [3, 4] and energy revolution [5, 6] has made great changes in the power industry and put forward more stringent requirements for the interaction and response of power supply and demand. To accurately evaluate the response ability of the power demand side [7], it is necessary to accurately identify the power load [8, 9], which is an important prerequisite for mastering the electricity consumption situation and electricity consumption behavior of users. In September 2020, China put forward the Dual Carbon Targets at the United Nations General Assembly [10, 11], that is, to achieve “Carbon Peak” by 2030 and achieve “carbon neutrality” by 2060. In December 2020, China further proposed measures to achieve the “Dual Carbon Targets” at the Climate Ambitious Summit [12]. It can be seen that energy saving and emission reduction and refined electricity consumption are the inevitable trends of the future development of the power industry. The research on the power load identification is helpful to provide users with efficient energy services and realize green and intelligent power consumption [13].
At present, many scholars have carried out in-depth research on the problem of power load identification, and a series of research results have emerged [14–16]. Aiming at the shortcomings of long training time and low recognition accuracy of existing algorithms, Huang et al. [17] proposed a load identification algorithm based on long-short-term memory-back propagation (LSTM-BP). On the basis of normalization and principal component analysis (PCA), the LSTM-BP power load identification model was constructed, and the model was verified by the REDD dataset. The results show that the proposed method has higher stability and accuracy than the traditional load identification algorithm. Considering that load identification is a key link in the power structure analysis, Yin et al. [18]proposed a similarity calculation method for spatial convex hull overlap rate, introduced transfer learning to identify unknown loads, and verified the performance of this method in the PLAID dataset. Since power load modeling has attracted more attention, Arif et al. [19] systematically reviewed the load modeling and identification technology and proposed the future research direction, focusing on the problems and new trends in load modeling and identification, so as to meet the growing interest of industry and academia. By combining underdetermined decomposition and feature filtering, Wu et al. [20] proposed a noninvasive load identification algorithm, which uses a two-step iterative shrinkage threshold algorithm to obtain the optimal solution. Then, according to the unique harmonic component of each power load, a feature filter is established to filter the decomposed power flow, so as to realize load identification. Based on the analysis of the actual measured current waveform of grid-connected equipment by power monitor, Beck et al. [21] proposed a practical power load identification method. A set of features was extracted by using the current physical component based on power theory decomposition, and the high-precision identification of power load was realized by using artificial neural network and the nearest neighbor search. The plug load accounts for one-third of the total energy consumption of commercial buildings. Therefore, Tekler et al. [22] proposed a near real-time identification method for plug load in office space. This method uses low-frequency power data (1/60 Hz), extracts power characteristics such as power, average power, and power delta, and finally realizes the identification accuracy of up to 93% with the help of the Bagging algorithm. In view of the limitations of the traditional household load identification method, Yu et al. [23] designed an event detector based on steady-state segmentation and a linear discriminant classifier group based on multi-feature global similarity to realize the accurate identification of household power load, and verified the effectiveness of the load identification method under the REDD dataset. In order to overcome the challenges in improving the accuracy of linear load and nonlinear load identification, Le et al. [24] proposed a new idea for power load identification, that is, generating a new transient feature based on Hilbert transform (HT), and then combining sequence to sequence LSTM (Seq2Seq LSTM) to achieve load identification. Subsequently, the feasibility of the proposed method was verified on BLUED dataset and PLAID dataset. For the purpose of improving the real-time performance of power load identification, Hamdi et al. [25] filtered the original power signal to reduce the data dimension, and then extracted the maximum value and its location, average value, final value, and area under the curve. Finally, the load identification was realized by template matching.
In general, the traditional identification methods of power load are mainly realized by the combination of time domain analysis, frequency domain analysis, feature extraction and classifier; therefore, how to select the appropriate features is crucial. In feature selection, swarm intelligence optimization algorithm can often show excellent performance, so it is widely used in feature selection. Whale optimization algorithm (WOA), as one of the swarm intelligence optimization algorithms, is widely used in many fields such as spectral analysis [26], satellite remote sensing [27], and medical auxiliary diagnosis [28]. By embedding the simulated annealing (SA) algorithm into the WOA, Mafarja et al. [29] proposed a new method for feature selection. The performance of the proposed method was evaluated on 18 standard benchmark datasets in the UCI database, which proved that the WOA had the ability to search the feature space and select the attributes with the largest amount of information for the classification task. To overcome the problem of high dimensionality of hyperspectral data, Kumar et al. [30] proposed a band screening method for hyperspectral images based on WOA, and verified the effectiveness of band screening on three benchmark datasets (Indian Pines, Pavia, and Salinas). To solve the problem of feature selection in high-dimensional data, Too et al. [31] proposed a new variant of WOA based on spatial boundary strategy to play the role of finding potential features from high-dimensional feature space. The effectiveness of this method is verified on 16 high-dimensional datasets collected by Arizona State University.
Considering these factors, this paper proposes a new feature selection method for the power load, that is, using WOA to realize the selection of the joint features in time domain and frequency domain, to ensure the accuracy of power load identification. Firstly, the authors briefly describe the flow chart of the power load feature selection method proposed in this paper, and then introduce the source of experimental data, the types of time domain, and frequency domain features of power load, as well as the application of WOA in the selection of power load joint features. Next, the authors analyze the experimental results. Based on the comparison of six kinds of electrical equipment power data, different classifiers are used to classify the original power data. Then, the time domain features and frequency domain features of the original power data are extracted and used for the identification of power load. On this basis, WOA is used to screen the joint features of power load, and the effectiveness of feature selection is verified by six classifiers. In addition, the authors also compare the proposed method with PCA for feature extraction to further verify the reliability of the model. Finally, the authors summarize this study and prospect future research work.
2. Proposed Methodology for Power Load Identification
The specific experimental process is shown in Figure 1. The authors select the line current, real power, reactive power, and apparent power of six typical electrical equipment such as clothes washer (CWE), kitchen dishwasher (DWE), kitchen convection wall oven (WOE), clothes dryer (CDE), kitchen fridge (FGE), and instant hot water unit (HTE) from The Almanac of Minutely Power dataset Version 2 (AMPds2) [32]. Firstly, the original power load dataset is obtained by dividing the original data into days, and then the invalid data (samples with all data of 0 in a day) are eliminated to obtain an effective and available power load dataset. Subsequently, the time domain features and frequency domain features of power load data are extracted, and the joint features (time domain and frequency domain) of power load are formed. Then, WOA is used to screen the characteristics of the power load. Finally, the selected feature information is used as the input of different classifiers to identify the type of power load, to verify the feasibility of WOA for feature selection of the power load.
2.1. Data Acquisition
To verify the effectiveness and feasibility of the method proposed in this paper, AMPds2 is used as the original data, which records the consumption of water, electricity, and natural gas in a residential building in Burnaby, Canada, in two years (730 days in total). Among them, the power-related measurement parameters mainly include line voltage, line current, line frequency, active power, reactive power, apparent power, etc. In particular, AMPds2 has been pre-cleaned at the time of release, which is very suitable for engineers and scientists in power, energy, construction, and other industries to test the performance of the model in a real environment.
Six kinds of typical electrical equipment are selected as the research object, and the line current (I), active power (P), reactive power (Q), and apparent power (S) of electrical equipment are taken as the input information. For 730 days of power data (sampled once per minute and the cumulative length of data is ), firstly, taking the time length of one day as the basic unit, divide the original power data (I, P, Q, and S) of each electrical equipment into 730 data, and combine the power data of each equipment into one data sample. Then, the samples with all power data of 0 (that is, the electrical equipment is not used) are screened out, and finally, the power datasets of six typical electrical equipment is formed. The power dataset includes 276 CWE samples, 250 DWE samples, 730 WOE samples, 349 CDE samples, 730 FGE samples, and 730 HTE samples, with a total of 3065 valid samples. In the process of power data analysis, the authors select 70% of the samples as the training set and 30% of the remaining samples as the test set, that is, the number of training set samples is 2142 and the number of test set samples is 923. In addition, when constructing the load identification model, the authors introduce 5-fold cross-validation to improve the reliability and effectiveness of the model.
2.2. Joint Features of Time Domain and Frequency Domain
The analysis of power signals can be mainly divided into time domain analysis [33], frequency domain analysis [34], and time-frequency domain analysis [35]. Considering the complexity of power load, this paper extracts the time domain features and frequency domain features of power data to form the joint features of power load, and on this basis, WOA is introduced for feature screening. In this paper, 16 time domain features and 13 frequency domain features are mainly used, of which the time domain features include 10 dimensional statistical indexes and 6 dimensionless statistical indexes. The calculation formulas of time domain features and frequency domain features are shown in Table 1.
In particular, the authors extract the time domain features and frequency domain features of power data (I, P, Q, and S) of electrical equipment, respectively, so that power load features can be obtained for each sample.
2.3. Description of WOA for Power Load Feature Screening
Inspired by the predatory behavior of humpback whales, Mirjalili et al. [36] proposed a whale optimization algorithm based on the simulation of the process of humpback whale swarm enclosure, hunting and attacking prey. The optimization algorithm has the advantages of less parameters, fast convergence speed, and strong global optimization ability. The basic idea of using WOA for the screening of the joint features of the power load time domain and the frequency domain is to determine the parameters to be optimized according to the power load time domain and frequency domain joint features screening problem, that is, the joint characteristics of the time domain and frequency domain of the power load, and the spatial position of each individual in the whale group contains a set of screening features. The fitness function is used to measure the spatial position of the individual, and the whale foraging strategy is used to continuously update the whale individual position until the best whale spatial position is obtained, which is the best set of screening features for the optimization problem.
The joint feature screening process of power load in the time domain and frequency domain is as follows: Step 1: Define the fitness function. Since WOA is a process to solve the minimum value, the classification error of the test set of the power load identification model is taken as the fitness function of this paper, that is, the objective function is: Where is the number of correctly predicted samples in the test set, and is the total number of samples in the test set. Step 2: Parameters initialization of WOA. A feature combination is randomly selected from the joint feature of time domain and frequency domain of power load as the initial whale position, and the parameters of WOA are set, including the number of groups , the maximum number of iterations , the selection of contraction bounding mechanism, and the probability of spiral position update ( is the random number on , and the initial value is set by random function). Step 3: Calculate the fitness value of each whale individual according to formula (1), find the best whale individual in the current group, and save the results. Step 4: When , if , update the spatial position of the individual of the current whale group according to formula. Where and are coefficient vectors, is the best spatial position of the current whale group, is the individual spatial position of the current whale group, and is the number of current iterations. In particular, the coefficient vectors and are calculated as follows: Where is a constant, and its value decreases linearly from 2 to 0; is a random vector, and its value range is . If , the individual position of the whale group is randomly selected from the current group, and the spatial position of the individual of the current whale group is updated according to the following formula: Where is the position randomly selected from the current whale population. Step 5: when , update the spatial position of the current whale population according to formula. where is the defined logarithmic spiral shape constant and is a random number in the range . Step 6: Calculate the classification error of the test set corresponding to each whale group individual as the fitness value, find the best whale group individual in the current group, and save the results. At the same time, judge whether the termination conditions are met. If so, go to step 7 and output the optimization result; otherwise, make t = 1, update A, B and C at the same time, and repeat steps 4 to 6 above. At the same time, determine whether to meet the termination conditions, if satisfied, then go to Step 7 and output the optimization results; otherwise, let , update , , , , and simultaneously, and repeat Step 4 to Step 6 above. Step 7: Output the individual fitness value of the optimal whale population and its spatial location , which is the best set of screening features.
2.4. Load Identification Model
For the purpose of classification of power load data, the basic classification algorithms, such as back propagation (BP) neural network, extreme learning machine (ELM), support vector machine (SVM), k-nearest neighbor (KNN), decision tree (DT), and naive Bayesian (NB), were used to construct the power load identification model, so as to realize the identification of power load types. In addition, in the process of building the power load identification model, the authors introduce 10-fold cross validation to increase the reliability and effectiveness of different classification models. And in the process of five independent experiments, we randomly divide the samples, so that the composition of the test set samples is different in each experiment.
3. Results and Discussion
3.1. Presentation of Raw Power Data
A total of 3065 effective samples of six types of typical electrical equipment are selected from the AMPds2. The authors select a day as an example to display the power data of six kinds of electrical equipment more intuitively. In the process of displaying the power load data, considering the difference of current and power data, the authors display these four power parameters, respectively, as shown in Figure 2.
It can be seen that the load curves of different electrical equipment have certain differences. For example, the current and functional parameters of WOE and CDE are relatively large, and their power curves are mainly concentrated in the use period. The power data of the refrigerator are relatively small, but their duration is very long. At the same time, the authors also note that the electrical parameters of some equipment also have a certain overlap, and the discrimination is not large. Therefore, it is necessary to use the signal analysis method to achieve more accurate identification.
3.2. Classification of Original Power Signals
In order to more intuitively understand the difference of power data of different electrical equipment, BP, ELM, SVM, KNN, DT, and NB are used as classifiers, and the original power data are used as input, that is, all power data (I, P, Q, and S) of a sample within a day are used as input. The classification accuracy and analysis time under different classifiers are observed, and five independent experiments are repeated. The relevant results are shown in Table 2. At the same time, the core parameters of the classifiers are described in the table.
First of all, the authors observe the accuracy. It can be found that the effect of the six classifiers is not particularly good. The average recognition rate of the five independent experiments using NB as the classifier is 73.44, while the average recognition rate of the other classifier is less than 50.00%, which shows that the accuracy of electrical equipment identification using original power data is low. In other words, the direct use of original power data cannot meet the requirements of accurate identification of electrical equipment. Subsequently, the authors focus on the analysis time. It can be seen that the analysis time of different classifiers is quite different. The analysis time of ELM is the shortest, with an average of only 0.4310 s, while the analysis time of KNN and DT is about 2 s. However, the analysis time of BP and SVM is much larger, which takes about 200 s. The most surprising thing is that the analysis time of NB reaches 4963.2828 s. The long analysis time of BP, SVM, and NB is mainly due to the fact that the dimension of original power data reaches . Considering the actual requirements of accuracy and speed for power load identification, the authors must process the original power data to achieve better identification performance.
3.3. Classification of Time Domain Features and Frequency Domain Features
According to the lack of low accuracy and slow analysis speed of the original power signal, the time domain features and frequency domain features of the original power signal are extracted according to the feature extraction method described in Table 1, so that the authors can obtain the time domain features, frequency domain features, and the joint features of the time domain and frequency domain of the power load sample. Since each load sample contains four electrical parameters, I, P, Q, and S, the dimension of the time domain feature of a sample is , the dimension of the frequency domain feature of a sample is , and the dimension of the joint feature of the time domain and frequency domain of a sample is . For the purpose of comparing the performance of different features in power load identification, the authors take the time domain features, frequency domain features, and joint features as the input vectors, feed them into different classifiers to construct the load identification models, and take the accuracy of the test set as the evaluation index of the model. Five independent experiments were carried out, and the mean and standard deviation of the five experiments were calculated. The classification performance of different classifiers was drawn, as shown in Figure 3.
In Figure 3, the authors show the mean value of classification accuracy in the form of a broken line graph, and the standard deviation in the form of an error band, so that the authors can clearly see the classification accuracy and stability of different classifiers. When observing the performance of different features, the authors find that the joint features of time domain and frequency domain have the best effect on power load identification, followed by the performance of power load identification using frequency domain features alone, and the performance of power load identification using time domain features alone is the worst. This shows that the frequency domain features of power data of different electrical equipment are more discriminating than the time domain features. At the same time, the joint features of time domain and frequency domain can effectively integrate their advantages and achieve the best identification effect. At the same time, the authors also find an interesting phenomenon. Although the classification effects of different features are quite different, the average classification accuracy of any feature exceeds 90.00%, which is much higher than the classification accuracy of the original power data. This shows that the feature extraction method described in Table 1 is effective and feasible for the identification of power load. Taking the joint feature as an example, when looking at the performance of different classifiers, the authors find that the classification accuracy of NB and SVM is high (the average accuracy is 99.78% and 99.72%, respectively), while the performance of ELM is poor (the average accuracy is 95.84%). In addition, the authors also noticed that the standard deviation of the classification accuracy of NB in the five independent experiments was 0, which maintained excellent classification stability. The standard deviation of the classification accuracy of five ELM experiments was 0.68%, and the stability was not as good as NB.
While recording the classification accuracy, the authors also analyze the classification time of different features under different classifiers, calculate the mean and standard deviation of the analysis time, and display it in a bar chart with error bars, as shown in Figure 4. Because of the large difference in the analysis time of different classifiers, the authors enlarge the analysis time of BP, ELM, SVM, KNN, and DT, so that the time cost of different classifiers can be more intuitively seen. First of all, the authors can find that the analysis time of frequency domain features is the shortest, followed by the analysis time of time domain features, and the analysis time of joint features is the longest. This is because the dimension of frequency domain features is the lowest ( dimensions), and the dimension of joint features is the highest ( dimensions). The more the feature dimension is, the more is the analysis time needed. Secondly, the authors focus on the time cost of different classifiers. It is obvious that the analysis time of NB is much longer than that of other species algorithms. When NB algorithm is used to analyze the joint features of power load, the average analysis time of five independent experiments is 54.7381 s, which is only 1.10% of the original power data analysis time, greatly reducing the time cost of the algorithm. The analysis time of the five algorithms is lower, in which the analysis time of SVM is 2.5546 s, and that of KNN is the shortest, only 0.0334 s. Thus, feature extraction of the original power data can not only improve the accuracy of power load identification but also reduce the time cost of the algorithm.
3.4. Selection of Joint Features by Using WOA
In order to extract the most effective features in the joint features, WOA is used to screen the features to further improve the analysis accuracy and reduce the analysis time. In the process of screening the joint features of power load using WOA, KNN classification model is established by using the selected feature data (KNN is selected as a classifier to reduce the time of WOA iterative screening features). The number of feature variables corresponding to the minimum classification error value is the final screening result. In particular, during feature selection using WOA, set the number of groups to 5, and set the maximum number of iterations to 100.
Because WOA is an algorithm to find the minimum value, in the process of load characteristic screening using WOA, we take the error rate as its objective function, that is, select the case with the minimum error rate. Figure 5 shows the variation trend of classification error of WOA in the process of screening load features. It can be seen from the figure that the classification error reaches the minimum value of 0.6529% after 29 iterations from the initial 1.8498%. At this time, the number of features screened is 15, which are , , , , , , , , , , , , , , and . In particular, the letters before the underscore represent a certain electrical parameter of the electrical equipment, and the specific features of the parameter behind the underscore. By observing these 15 features, the authors find that these features are mainly concentrated in the amount indicating the dispersion or concentration of the spectrum.
3.5. Classification of WOA Screening Features
The effectiveness and reliability of the features screened by WOA are further verified. The selected feature data are used as the input information of power load identification. The identification models of power load types are constructed by using BP, ELM, SVM, KNN, DT, and NB, respectively. Five independent experiments are carried out, and the mean and standard deviation of the six classification models are counted and calculated. The classification accuracy and analysis time are shown in Figure 6.
The authors first pay attention to the accuracy of the classification model. It can be seen that when the features selected by WOA are used for power load identification, no matter which classifier has a very good performance, the ELM with the worst classification effect can also achieve an average recognition rate of 97.94%, and the average recognition accuracy of the five classification algorithms is about 99.00% (in which NB can achieve 100.00% recognition accuracy of power load when it is used as the classifier). Therefore, WOA can effectively screen out the most effective features for power load identification in the joint features of domain and frequency domain. In addition, the authors observe the analysis time. The analysis time of the six classifiers has decreased to varying degrees, which is mainly due to the fact that WOA carries out load feature selection and reduces the dimension of data. Among them, the analysis time of KNN is the shortest, only 0.0107 s, while the analysis time of NB is still the longest, 6.3631 s. At this time, the analysis time of NB classifier is 11.62% of the analysis time of the joint feature data, and 0.13% of the analysis time of the original power data. In summary, using WOA to filter the features of power load can not only effectively improve the accuracy of power load identification but also greatly reduce the time cost of the algorithm model.
For the purpose of showing the effectiveness of power load identification more intuitively, the authors take DT classifier as an example and use confusion matrix to show the identification results of power load, as shown in Table 3. By observing the data in the table, the authors can clearly see the forecast of each electrical equipment. Among the 83 CWE samples in the test set, 80 samples can accurately predict the category, 1 sample is wrongly predicted as DWE, and 2 samples are wrongly predicted as FGE. Among 75 DWE samples in the test set, 73 samples can accurately predict the category, and 2 samples are wrongly predicted as CWE. Among 220 WOE samples in the test set, 218 samples can accurately predict the category, 1 sample is wrongly predicted as DWE, and 1 sample is wrongly predicted as CDE. Among the 105 CDE samples in the test set, 102 samples can accurately predict the category, and 3 samples are wrongly predicted as WOE. All 220 FGE samples in the test set can accurately predict the correct category. Among 220 HTE samples in the test set, 218 samples can accurately predict the category, and 2 samples are wrongly predicted as WOE.
3.6. Comparison with Traditional Strategies
As is known to all, PCA is a common and effective feature extraction and dimension reduction method, which is widely used in pattern recognition. Therefore, it is necessary to compare the feature selection idea proposed in this paper with the feature extraction effect of PCA. In particular, before the process of PCA, the authors normalize the feature data, namely, the original feature is normalized to the interval [0,1]. At the same time, in the process of PCA, the authors set the cumulative contribution threshold to 95%. Subsequently, the obtained principal component information was fed into six classifiers to construct the power load identification model, and five independent experiments were carried out. The identification performance of PCA combined with different classifiers for power load was statistically analyzed, as shown in Figure 7. To see the results of the five experiments more clearly, the authors enlarged some details.
It can be seen from the graph that the performance of PCA combined with different classifiers for power load identification is quite different. For example, the average recognition accuracy of PCA combined with BP for load identification is only 36.62%, and the average classification accuracy of PCA combined with ELM or SVM is only about 68%. The recognition effect of PCA combined with NB classifier is the best, which can reach the average recognition accuracy of 94.03%. However, when the authors compared the results of feature selection with WOA, the authors found that the average recognition accuracy (94.03%) of the power load identification strategy of even PCA with NB with the best classification effect was 3.91% lower than that of the ELM classifier with the worst load feature classification effect selected by WOA (97.94%). Therefore, compared with the traditional dimension reduction strategy of PCA, WOA has more excellent performance in feature selection of power load.
4. Conclusions
Accurate classification of power load is the premise of demand side response capability evaluation, which is of great significance for demand side management. Power load identification is essentially a supervised learning and forecasting process. How to obtain the characteristics of load identification is very important to ensure the accuracy of power load identification. Therefore, a feature selection method based on WOA is proposed in this paper to select the joint features of power load in time domain and frequency domain. The electrical measurement data of six typical electrical equipment are selected to form the original power load identification dataset. Firstly, the time domain features and frequency domain features of the power load data are extracted. And, on the basis of extracting the time domain features and frequency domain features of power load, WOA is used to further screen the most useful feature combination for power load identification. Finally, the selected feature information is used as input to verify the effect of WOA on the joint features of power load under BP, ELM, SVM, KNN, DT, and NB classifiers. The research results show that, compared with the original power data, extracting the time domain and frequency domain features of power data can improve the accuracy of power load identification. Compared with the time domain features of power load, the frequency domain features of power load are more discriminating, and the effect of joint features (the combination of time domain and frequency domain) is better. Using WOA to screen the joint features, 15 feature attributes that are most helpful for power load identification can be selected from the original 116 power load features. The average classification accuracy of the 15 feature attributes under the six classifiers is 99.16%, and the average analysis time is 1.4531 s. In addition, compared with the traditional PCA feature extraction strategy, WOA has more excellent performance in feature selection of power load. In summary, WOA is used to screen the joint features of power load, and the accurate identification of power load is realized, which provides a basis for further developing the differentiated demand side response strategy and fine evaluation of demand side response potential. At the same time, it is of great significance for intelligent dispatching of the power system and improving the economic operation of the power system.
At present, the research work in this paper still has certain limitations, mainly reflected in the following two aspects. On the one hand, the current load data source is residential load, and the load type is relatively small. In the future, the authors will introduce industrial load and commercial load to further enrich the load type. On the other hand, the power load feature selection method proposed in this paper is based on the extraction of time domain features and frequency domain features of the load, and the feature extraction is a cumbersome process. Therefore, in the future, the authors will explore the feasibility of power load identification from the perspective of power signal decomposition.
Data Availability
The data used to support the findings of this study are available from the corresponding author upon request. In addition, data can be obtained by visiting http://ampds.org/.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
This work was supported by the Key Projects of Natural Science Research in Anhui Universities (grant numbers KJ2021A0470, KJ2021A0471); the University-level Key Projects of Anhui University of Science and Technology (grant number xjzd2020-06); the Talent Introduction Fund of Anhui University of Science and Technology (grant number 13200404); the Young Talent Project of Anhui University of Science and Technology (grant number 2020023); the National Key R&D Program of China (grant number 2020YFB1314100); the Energy Internet Joint Fund of Anhui Province (grant number 2008085UD06); and the Anhui Science and Technology Major Project (grant number 201903a07020013).