Household electricity non-intrusive load monitoring method based on load feature library

Abstract

The invention belongs to the technical field of power load monitoring, and discloses a household electricity non-intrusive load monitoring method based on a load feature library, and the method comprises the steps: obtaining historical load data related to the historical operation state of N household appliances; according to the historical power data, obtaining a load characteristic of a switching event of each household electrical appliance; dividing the corresponding N household appliances into M types according to the N load characteristics, and calculating the type characteristics of the switching event of each type of household appliances; constructing a feature library according to the category features, and establishing a load decomposition model based on the feature library; acquiring real-time load data related to real-time operation of N household appliances, taking the real-time load data as input of the load decomposition model, performing optimization solution on the load decomposition model by using a dynamic integer programming algorithm, and outputting to obtain real-time operation states of M types of household appliances; wherein 1 < = M < = N.

Description

Technical Field

[0001] The present invention belongs to the technical field of power load monitoring, and particularly relates to a non-intrusive load monitoring method for household electricity consumption based on a load feature library. Background Art

[0002] At present, load monitoring is mainly divided into two categories: intrusive load monitoring and non-intrusive load monitoring;

[0003] Intrusive load monitoring can well provide the power consumption habits and demands of users for power companies by placing sensors on electrical equipment, and obtain very accurate load data of them. However, this method is costly and involves user privacy issues. Non-intrusive load monitoring only uses the total load characteristics as information, and monitors the power consumption of each internal electrical equipment by analyzing the total load data, which can well overcome the disadvantages of intrusive load monitoring.

[0004] The main steps of non-intrusive load monitoring are: data acquisition, data preprocessing, and load decomposition. Among them, data preprocessing includes modules such as outlier detection, sampling rate conversion, event detection, and feature extraction. The two most important steps for non-intrusive load monitoring are feature extraction and load decomposition. Feature extraction aims to extract the characteristics of different electrical appliances during operation. Load characteristics are various types of information that can reflect the characteristics of electrical equipment and are used to distinguish it from other equipment, also known as device fingerprints or device signatures. Through these characteristics, the electrical appliances being used by the user can be identified; load decomposition is to decompose the total power into the powers of each electrical equipment. The differences between existing different non-intrusive load monitoring schemes lie in the different load feature extraction and load decomposition algorithms. However, the original intention of using non-intrusive load monitoring is to protect user privacy, and the method is to rely only on the total load data. However, existing non-intrusive load monitoring schemes need to use the electrical load data of each electrical appliance to extract features, and then perform load decomposition to obtain the usage conditions of each electrical appliance, which is contrary to the original intention. Summary of the Invention

[0005] In view of this, to solve the problems raised in the above background art, the purpose of the present invention is to provide a non-intrusive load monitoring method for household electricity consumption based on a load feature library.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] A non-intrusive load monitoring method for household electricity consumption based on a load feature library, comprising:

[0008] Obtaining historical load data related to the historical operating status of household appliances;

[0009] Obtaining the load characteristics of switching events of the household appliances according to the historical power data;

[0010] Construct a feature library according to the load characteristics, and establish a load decomposition model based on the feature library;

[0011] Obtain real-time load data related to the real-time operation of household appliances, use the real-time load data as the input of the load decomposition model, and use the dynamic integer programming algorithm to optimize and solve the load decomposition model, and output the real-time operation status of household appliances.

[0012] A non-intrusive load monitoring method for household electricity based on a load feature library, including:

[0013] Obtain historical load data related to the historical operation status of N household appliances;

[0014] Obtain the load characteristics of each household appliance when a switching event occurs according to the historical power data;

[0015] Divide the corresponding N household appliances into M categories according to the N load characteristics, and calculate the category characteristics of the switching events occurring in each category of household appliances;

[0016] Construct a feature library according to the category characteristics, and establish a load decomposition model based on the feature library;

[0017] Obtain real-time load data related to the real-time operation of N household appliances, use the real-time load data as the input of the load decomposition model, and use the dynamic integer programming algorithm to optimize and solve the load decomposition model, and output the real-time operation status of M categories of household appliances;

[0018] Among them, 1 ≤ M ≤ N.

[0019] Preferably, the step of dividing the corresponding N household appliances into M categories according to the N load characteristics includes:

[0020] Use the K-means++ clustering algorithm to cluster the N load characteristics;

[0021] Divide the N household appliances into M categories according to the clustering result.

[0022] Preferably, the step of calculating the category characteristics includes:

[0023] After clustering, obtain the load characteristics corresponding to the cluster center of each category;

[0024] Mark the load characteristics corresponding to the cluster center as the category characteristics of the switching events occurring in the household appliances of this category.

[0025] Preferably, both the historical load data and the real-time load data include the total active power and the total reactive power.

[0026] Preferably, the load characteristics include the active power increment and the reactive power increment.

[0027] Preferably, the steps of obtaining the load characteristics include:

[0028] Calculating the target active power increment and the target reactive power increment at time T and time T + 1 according to the historical load data;

[0029] Judging whether the target active power increment exceeds a threshold. If yes, it is determined that a target switching event occurs at time T + 1, and the target active power increment and the target reactive power increment are marked as the load characteristics of the target switching event.

[0030] Preferably, the load decomposition model is expressed as:

[0031] ;

[0032] In the formula, S and C are respectively time series composed of the total active power and the total reactive power within the target decomposition time period, P and Q are respectively the feature libraries corresponding to the active power and the reactive power, and W is the output matrix of the optimization solution.

[0033] Preferably, the output matrix W is expressed as:

[0034] ;

[0035] Wherein, Indicates that the j-th electrical appliance is in the on state at the i-th moment, Indicates that the j-th electrical appliance is in the off state at the i-th moment.

[0036] Preferably, the optimization solution process further includes:

[0037] Constructing a constraint function,

[0038] Applying the constraint function to the optimization solution process of the load decomposition model;

[0039] The constraint function is expressed as ; Wherein, Is the j-th row of the matrix W, Is the Manhattan norm, and a is a constraint constant.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] Based on the load data of total active power and reactive power, obtain the active power increment and reactive power increment when switching events of a series of electrical appliances occur, and use these as the load characteristics when the electrical appliances have switching events. Based on this, construct a load characteristic library, and decompose the total load under the real-time operation of the electrical appliances to match the load characteristic library using a dynamic integer programming algorithm to obtain the usage conditions of each electrical appliance.

[0042] Furthermore, cluster the load characteristics of each electrical appliance that has been extracted, thereby obtaining the category characteristics when switching events of various types of electrical appliances occur, and use this to divide all household electrical appliances into several categories according to the category characteristics. During the load decomposition process, only analyze the electricity usage patterns of electrical appliances of each type, avoiding excessive collection of user information, and thus effectively protecting the privacy of users. Description of the Drawings

[0043] Figure 1 Flowchart of the non-intrusive load monitoring method for household electricity based on the load characteristic library of the present invention;

[0044] Figure 2 Active power and reactive power load curves for the experimental example

[0045] Figure 3 Display chart of the clustering processing results for the experimental example;

[0046] Figure 4 Comparison chart of the actual active power and the reconstructed active power for a certain day in the experimental example. Detailed Implementation Manner

[0047] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0048] Embodiment 1

[0049] A non-intrusive load monitoring method for household electricity based on a load characteristic library, including:

[0050] Obtain historical load data related to the historical operating status of household electrical appliances;

[0051] Obtain the load characteristics of the household electrical appliances when switching events occur according to the historical power data;

[0052] Construct a feature library according to the load characteristics, and establish a load decomposition model based on the feature library;

[0053] Obtain real-time load data related to the real-time operation of household appliances, use the real-time load data as the input of the load decomposition model, and use the dynamic integer programming algorithm to optimize and solve the load decomposition model, and output the real-time operation state of household appliances.

[0054] Embodiment 2

[0055] A non-intrusive load monitoring method for household electricity based on a load feature library, including:

[0056] S1. Obtain historical load data (including total active power and total reactive power) related to the historical operation states of N household appliances.

[0057] S2. Obtain the load characteristics of each household appliance for switching events according to the historical power data;

[0058] Specifically in this step:

[0059] Calculate the target active power increment and target reactive power increment at time T and time T+1 according to the historical load data;

[0060] Judge whether the target active power increment exceeds the threshold. If so, it is determined that a target switching event occurs at time T+1, and mark the target active power increment and target reactive power increment as the load characteristics of the target switching event.

[0061] S4. Use the K-means++ clustering algorithm to cluster the N load characteristics;

[0062] K-means++ is an improved K-means clustering algorithm, which can better initialize the cluster centers, thereby improving the accuracy and stability of the clustering results. In the K-means algorithm, all initial cluster centers are usually randomly selected, which may lead to unstable clustering results because the positions of the initial cluster centers may affect the final clustering results. The K-means++ algorithm selects the initial cluster centers in a heuristic way, making the initial cluster centers farther away from each other, thereby improving the accuracy and stability of the clustering results.

[0063] S4. Divide the N household appliances into M categories according to the clustering result, and obtain the load characteristics corresponding to the cluster centers of each category, and mark the load characteristics corresponding to the cluster centers as the category characteristics of the switching events of the household appliances of this category;

[0064] S5. Construct a feature library according to the category characteristics (the feature library includes the active power and reactive power increments generated when M types of events occur, that is, category characteristics), and establish a load decomposition model based on the feature library;

[0065] The load decomposition model is expressed as:

[0066] ; where S and C are time series composed of the total active power and total reactive power during the target decomposition time period, P and Q are feature libraries corresponding to the active power and reactive power respectively, and W is the output matrix of the optimization solution.

[0067] The output matrix W is expressed as:

[0068] ; where indicates that the j-th electrical appliance is in the on state at the i-th moment, indicates that the j-th electrical appliance is in the off state at the i-th moment.

[0069] S6. Obtain the real-time load data related to the real-time operation of N household appliances, use the real-time load data as the input of the load decomposition model, and use the dynamic integer programming algorithm to optimize and solve the load decomposition model, and output the real-time operation states of M types of household appliances;

[0070] where 1 ≤ M ≤ N.

[0071] In addition, since there is a certain similarity between the switch states of household appliances and the previous moment, a constraint function is constructed and applied to the optimization and solution process of the load decomposition model; the constraint function is expressed as:

[0072] ; where is the j-th row of the matrix W, is the one-norm of the vector (also known as the Manhattan norm, and the calculation method is the sum of the absolute values of each element of the vector), and a is the constraint constant.

[0073] In the present invention, the K-means++ clustering algorithm can be replaced by a similar algorithm, and the dynamic integer programming algorithm can also be replaced by a similar algorithm (such as a combinatorial optimization algorithm).

[0074] For the above-mentioned non-intrusive load monitoring method of household electricity based on the load feature library, the technical solution will be introduced in detail below in combination with the AMPds2 dataset. This dataset contains the complete usage data of electricity, water and natural gas of a household in Burnaby for two years with a granularity of 1 minute. The active power and reactive power in a certain week (a total of 10080 minutes) in 2013 in this dataset are selected for experiments:

[0075] As Figure 2 shows the active power and reactive power load curves for this selected week. The figure shows that the load curves are constantly fluctuating, which is caused by electrical switching events.

[0076] According to Figure 2 the provided data, calculate the target active power increment and target reactive power increment at time T and time T+1. When the target active power increment exceeds the set threshold (since the mutation of active power is accompanied by the occurrence of reactive power mutation, only setting the inspection of one threshold can well monitor the occurrence of events), it is determined that there is a target switching event at time T+1, and the target active power increment and target reactive power increment are marked as the load characteristics of the target switching event.

[0077] Use the K-means++ clustering algorithm to cluster N load characteristics, and the clustering result is as Figure 3 shown. The set number of clustering centers is M, and thus the active power and reactive power characteristic libraries of M types of electrical appliances during stable operation can be obtained.

[0078] Obtain the real-time load data related to the real-time operation of N household appliances, and input the real-time load data into the load decomposition model ( ). Use the dynamic integer programming algorithm to optimize and solve the load decomposition model to obtain the output matrix W that can reflect the real-time operation status of various household appliances.

[0079] Recombine the active power load curve according to the real-time operation status of various household appliances output, and obtain a comparison graph as Figure 4 shown.

[0080] Compare the above experiments, and use the mean absolute error and coefficient of determination to evaluate the solution accuracy of the load decomposition model. The specific introductions of the two evaluation indicators are as follows:

[0081] (1) Mean absolute error (MAE): It represents the average absolute difference between the predicted value and the true value. Since the absolute value is taken, the positive and negative signs of each error are ignored, ensuring the non-negativity of all differences.

[0082] The expression is: ; where n is the sample length, is the predicted value of the model, is the corresponding true value.

[0083] (2) Coefficient of determination : It is a commonly used statistic in regression analysis, which measures the degree of explanation of the independent variable to the dependent variable. Simply put, it measures the goodness of fit of the model to the data, that is, the proportion of the variability of the dependent variable that the model can explain in the total variability. The closer the coefficient of determination is to 1, the better the fitting degree of the data.

[0084] The expression is: ; ; ; wherein, is the sum of squared residuals (representing the difference between the predicted value and the actual value), total sum of squares (representing the total variability of the dependent variable data), is the mean of the actual values.

[0085] According to the above calculation formula and Figure 4 the data shown, the mean absolute error is calculated to be 96.5348, and the coefficient of determination is 0.98287, indicating that the load decomposition model and the non-intrusive load monitoring method provided by the present invention have good prediction and monitoring accuracy.

[0086] In summary, the main difference between the present invention and other methods is that the present invention only relies on active power and reactive power data and does not use intrusive load monitoring data to extract the characteristics of each electrical appliance. Further, household appliances are classified into several levels according to the load level, and then by decomposing the total active power and reactive power data, the usage conditions of electrical appliances at different levels are obtained, avoiding excessive collection of user information, and thus effectively protecting the privacy of users.

[0087] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A non-intrusive load monitoring method for household electricity based on a load feature library, characterized in that, It includes the following steps: Obtain historical load data related to the historical operating status of household appliances; Obtain the load characteristics of switching events occurred to the household appliances according to the historical power data; Construct a feature library according to the load characteristics, and establish a load decomposition model based on the feature library; Obtain real-time load data related to the real-time operation of household appliances, use the real-time load data as the input of the load decomposition model, and optimize and solve the load decomposition model by using the dynamic integer programming algorithm, and output the real-time operation status of the household appliances.

2. The non-intrusive load monitoring method for household electricity based on a load feature library according to claim 1, characterized in that: Obtain historical load data related to the historical operating status of N household appliances; Obtain the load characteristics of switching events occurred to each household appliance according to the historical power data; Divide the corresponding N household appliances into M categories according to the N load characteristics, and calculate the category characteristics of switching events occurred to the household appliances in each category; Construct a feature library according to the category characteristics, and establish a load decomposition model based on the feature library; Obtain real-time load data related to the real-time operation of N household appliances, use the real-time load data as the input of the load decomposition model, and optimize and solve the load decomposition model by using the dynamic integer programming algorithm, and output the real-time operation status of the M categories of household appliances; wherein, 1 ≤ M ≤ N.

3. The non-intrusive load monitoring method for household electricity based on a load feature library according to claim 2, characterized in that, The step of dividing the corresponding N household appliances into M categories according to the N load characteristics includes: Performing clustering processing on the N load characteristics by using the K-means++ clustering algorithm; Dividing the N household appliances into M categories according to the clustering processing result.

4. The non-intrusive load monitoring method for household electricity based on a load feature library according to claim 3, characterized in that, The step of calculating the category characteristics includes: Obtaining the load characteristics corresponding to the cluster center of each category after clustering processing; Marking the load characteristics corresponding to the cluster center as the category characteristics of switching events occurred to the household appliances in this category.

5. A non-intrusive load monitoring method for household electricity based on a load feature library according to claim 1, 2, 3 or 4, characterized in that: Both the historical load data and the real-time load data include total active power and total reactive power.

6. The non-intrusive load monitoring method for household electricity based on a load feature library according to claim 5, characterized in that: The load characteristics include active power increment and reactive power increment.

7. The non-intrusive load monitoring method for household electricity based on a load feature library according to claim 6, wherein The step of obtaining the load characteristics includes: Calculating the target active power increment and target reactive power increment at time T and time T+1 according to the historical load data; Judging whether the target active power increment exceeds the threshold. If yes, it is determined that a target switching event occurs at time T+1, and mark the target active power increment and target reactive power increment as the load characteristics of the target switching event.

8. A non-intrusive load monitoring method for household electricity based on a load feature library according to claim 7, characterized in that, The load decomposition model is expressed as: ； In the formula, S and C are time series composed of total active power and total reactive power respectively within the target decomposition time period, P and Q are the feature libraries corresponding to active power and reactive power respectively, and W is the output matrix of the optimization solution.

9. A non-intrusive load monitoring method for household electricity based on a load feature library according to claim 8, characterized in that The output matrix W is expressed as: ； Among them, indicates that the j-th electrical appliance is in the on state at the i-th moment, indicates that the j-th electrical appliance is in the off state at the i-th moment.

10. A non-intrusive load monitoring method for household electricity based on a load feature library according to claim 9, characterized in that It also includes: Constructing a constraint function, Applying the constraint function to the optimization solution process of the load decomposition model; The constraint function is expressed as ; where is the j-th row of matrix W, is the Manhattan norm, and a is a constraint constant.

Images