CN111553444A - A load identification method based on non-intrusive load terminal data - Google Patents

Abstract

The invention discloses a load identification method based on non-invasive load terminal data, which comprises the following steps: sampling the load data; carrying out mean shift clustering on historical electricity utilization data of various electric appliances of a user to obtain different clusters of a clustered electric appliance data set of the user, extracting load identification characteristics of the electric appliances from the electric appliance clusters, thereby identifying the historical load type loads and establishing a multivariate Gaussian distribution model of the operation characteristics of the electric appliances; on the other hand, load operation data is used as input characteristics, and the probability of occurrence of each cluster corresponding to the operation characteristics of the electric appliance is calculated by a naive Bayes algorithm; and taking a cluster with the maximum probability as the actual category of the electric appliance, and taking the electric appliance type of the cluster as the electric appliance type of the electric appliance. According to the invention, through carrying out cluster modeling on the user historical load operation data sent from the non-invasive load terminal data, the re-identification of the cloud platform on the user load is realized, and the comparison and correction of the electrical appliance attributes sent from the non-invasive terminal can be effectively carried out.

Description

A load identification method based on non-intrusive load terminal data

technical field

The invention belongs to the technical field of intelligent power consumption, and in particular relates to a load identification method based on non-invasive load terminal data.

Background technique

Smart electricity consumption makes residential load a relatively controllable resource, and realizes load transfer and peak shaving and valley filling through means such as direct load control and time-of-use electricity pricing. Load monitoring is the core link of smart electricity consumption. Smart meters are used to analyze the internal load components and load characteristics of users, and obtain detailed information on electricity consumption behavior and electrical energy consumption of electrical appliances. Load monitoring is divided into intrusive load monitoring (Intrusive Load Monitoring, ILM) and non-intrusive load monitoring (Non-intrusive Load Monitoring, NILM). In order to perform online monitoring of all electrical equipment of users in the existing monitoring system, many sensors and data transmission devices for inductive measurement need to be installed in the home, which not only increases the installation cost and maintenance cost, but also makes this "intrusive" The installation method and maintenance management of the electricity will interfere with the production and life of users, and will reduce the satisfaction of users with energy management services. Therefore, it is urgent to improve the level of load monitoring technology in intelligent electricity consumption. The non-intrusive method is to install the monitoring equipment at the user's power entrance, analyze the user's power consumption information through the load identification algorithm, and learn the power consumption of each device inside the user, which greatly simplifies the hardware structure and reduces costs, and is suitable for a large number of scattered users. This mode brings benefits to the power grid and users, and has become a research hotspot at home and abroad.

The process of using non-intrusive load monitoring to realize the user's power consumption behavior analysis can be divided into three parts: load classification, power consumption behavior decomposition and advanced application. Firstly, the household electrical equipment is analyzed by feature extraction and load classification, and then the electricity consumption information of the electrical equipment is counted by the power consumption behavior decomposition algorithm, including the power consumption, the type of start and stop appliances, the electricity cost, the start and stop time, etc. , considering the complexity of non-intrusive load monitoring calculations, it generally runs on the cloud platform, and finally after the cloud analysis, the monitored power consumption information of each electrical equipment is fed back to the user, which is convenient for the user to manage the home energy and participate in the power grid. Interaction; on the other hand, it provides services for grid companies or other management departments to formulate incentive policies such as electricity prices and demand response measures.

To sum up, the breakthrough and industrialization of NILMD technology are of great significance to the development of smart electricity consumption and the improvement of energy conservation and emission reduction benefits. However, the power load feature identification technology of NILMD technology in the cloud platform needs to be further studied.

SUMMARY OF THE INVENTION

Purpose of the invention: In order to overcome the deficiencies in the prior art, a load identification method based on non-intrusive load terminal data is provided, which realizes the re-identification of the user load by the cloud platform, and can effectively identify the electrical properties sent by the non-intrusive terminal. Make a comparison correction.

Technical solution: In order to achieve the above purpose, the present invention provides a load identification method based on non-intrusive load terminal data, comprising the following steps:

S1: Sampling the load data through the non-intrusive load terminal;

S2: Organize the load data, which specifically includes the following two aspects:

On the one hand, the mean-shift clustering is performed on the historical power consumption data of various electrical appliances of the user to obtain different clusters of the user electrical appliance data set after clustering, and the load identification features of the electrical appliances are extracted from the electrical appliance clusters, so as to analyze the historical load type load. identify;

On the other hand, according to the user's electricity consumption characteristics of each sample point in the cluster, a multivariate Gaussian distribution model of the electrical appliance operating characteristics is established, and the load operating data is used as the input feature, and the naive Bayesian algorithm is used to calculate the electrical appliance operating characteristics corresponding to the occurrence of each cluster. probability;

S3: Take a cluster with the highest probability as the actual category of the appliance, obtain the appliance type to which the cluster belongs, and use the appliance type of the cluster as the appliance type of the appliance.

Further, the specific steps of mean shift clustering in the step S2 are as follows:

A1: Determine the sliding window radius r, and start clustering with a randomly selected center point C and a circular sliding window with a radius of r;

A2: In each iteration process, calculate the mean value in the new sliding window, that is, the new center point of the window, and the number of points in the sliding window is the density in the window;

A3: Continue to move the sliding window according to the mean value until the density in the circle no longer increases;

A4: Filter the sliding windows generated in steps A2 to A4. When multiple windows overlap, retain the window containing the most points, and perform clustering according to the sliding window where the data points are located;

A5: Save the centroid coordinates of each data set as the standard feature value of the category, and count the distribution of each feature in the group as the distribution range of electrical features.

Further, the historical power consumption data of the electrical appliance in the step S2 includes the electrical appliance's total electrical power consumption, peak power, start and stop times, on time and running time.

Further, the load identification features of the electrical cluster in the step S2 include the average load running time, the load average power, and the running time distribution.

Further, the determination process of the radius parameter r of the circular sliding window in the step S2 is:

Given a dataset P={p(i); i=0,1,...n}, for any point P(i), compute the subset S={p(1) from point P(i) to set D, The distances between all points in p(2),...,p(i-1),p(i+1),...,p(n)}, the distances are sorted in ascending order, assuming the sorted distances The set is D={d(1),d(2),...,d(k-1),d(k),d(k+1),...,d(n)}, then d(k) is called For the k-distance, the k-distance is calculated for each point p(i) in the clustering set, and finally the k-distance set of all points is obtained E={e(1),e(2),...,e(n )}, according to the obtained k-distance set E of all points, sort the set E in ascending order to obtain the k-distance set E', fit a change curve of k-distance in the sorted E' set, according to the change The graph determines the value of the k-distance corresponding to the position where the sharp change occurs as the value of the radius r.

Further, the process of establishing the multivariate Gaussian distribution model of the operating characteristics of the electrical appliance in the step S2 is:

B-1: Preliminary data processing:

According to the different clusters of the clustered consumer electrical appliance data set, the user electricity consumption characteristics of the points in each cluster are respectively read;

B-2: Calculate the feature mean μ of the sample:

Among them, k is the total number of feature i in the electrical cluster, and μ _i is the feature mean of feature i in the cluster;

B-3: Establish the covariance matrix Σ of the characteristic distribution of electricity consumption, X is the eigenvector set of the sample, m is the number of samples,

The multivariate Gaussian distribution model of the user's electricity consumption behavior is obtained.

Further, in the step S2, the specific calculation steps for calculating the occurrence probability of each cluster corresponding to the operating characteristics of the electrical appliance by the Naive Bayes algorithm are as follows:

C-1: Determine the operating characteristics of the sample set and collect sample data;

C-2: For training samples, the conditional probability of its features is calculated for each category, that is, the probability of occurrence of the corresponding feature value of each feature;

C-3: target classification, read the feature value X of the target, calculate its probability p(X|C _i ) in each category C _i , and take the largest item as the target class.

Beneficial effect: Compared with the prior art, the present invention performs cluster modeling on the user's historical load operation data sent by the non-intrusive load terminal data, and then analyzes the type of the user's actual electrical appliances, thereby realizing the cloud platform's re-evaluation of the user's load. It can effectively compare and correct the electrical properties sent by the non-intrusive terminal.

Description of drawings

Fig. 1 is the system flow chart of the method of the present invention;

FIG. 2 is a power diagram of a case for testing the method of the present invention.

Detailed ways

The present invention will be further illustrated below in conjunction with the accompanying drawings and specific embodiments.

As shown in FIG. 1, the present invention provides a load identification method based on non-intrusive load terminal data, including the following steps:

S1: Sampling the load data through the non-intrusive load terminal;

S2: Organize the load data through the cloud platform, which specifically includes the following two aspects:

On the one hand, the mean-shift clustering is performed on the historical power consumption data of various electrical appliances of the user to obtain different clusters of the user electrical appliance data set after clustering, and the load identification features of the electrical appliances are extracted from the electrical appliance clusters, so as to analyze the historical load type load. Identification, the load identification features of the electrical cluster include the average running time of the load, the average power of the load, and the distribution of running time;

On the other hand, according to the user's electricity consumption characteristics of each sample point in the cluster, a multivariate Gaussian distribution model of the electrical appliance operating characteristics is established, and the load operating data is used as the input feature, and the naive Bayesian algorithm is used to calculate the electrical appliance operating characteristics corresponding to each cluster. probability of occurrence;

S3: Take a cluster with the highest probability as the actual category of the appliance, obtain the appliance type to which the cluster belongs, and use the appliance type of the cluster as the appliance type of the appliance.

The historical power consumption data of the electrical appliance in this embodiment includes the electrical appliance's total electrical power consumption, peak power, start and stop times, on time, and running time, and the like.

The specific steps of mean-shift clustering in step S2 in this embodiment are as follows:

A1: Determine the sliding window radius r, and start clustering with a randomly selected center point C and a circular sliding window with a radius of r;

A2: In each iteration process, calculate the mean value in the new sliding window, that is, the new center point of the window, and the number of points in the sliding window is the density in the window;

A3: Continue to move the sliding window according to the mean value until there is no direction that can accommodate more points in the kernel, that is, until the density in the circle no longer increases;

A4: Filter the sliding windows generated in steps A2 to A4. When multiple windows overlap, retain the window containing the most points, and perform clustering according to the sliding window where the data points are located;

A5: Save the centroid coordinates of each data set as the standard feature value of the category, and count the distribution of each feature in the group as the distribution range of electrical features.

Among them, the determination process of the radius parameter r of the circular sliding window is:

Given a dataset P={p(i); i=0,1,...n}, for any point P(i), compute the subset S={p(1) from point P(i) to set D, The distances between all points in p(2),...,p(i-1),p(i+1),...,p(n)}, the distances are sorted in ascending order, assuming the sorted distances The set is D={d(1),d(2),...,d(k-1),d(k),d(k+1),...,d(n)}, then d(k) is called For the k-distance, the k-distance is calculated for each point p(i) in the clustering set, and finally the k-distance set of all points is obtained E={e(1),e(2),...,e(n )}, according to the obtained k-distance set E of all points, sort the set E in ascending order to obtain the k-distance set E', fit a change curve of k-distance in the sorted E' set, and then draw The curve is drawn, and through observation, the value of the k-distance corresponding to the position where the sharp change occurs is determined as the value of the radius r.

In the present embodiment, the process of establishing the multivariate Gaussian distribution model of the operating characteristics of the electrical appliance in step S2 is as follows:

B-1: Preliminary data processing:

According to the different clusters of the clustered consumer electrical appliance data set, the user electricity consumption characteristics of the points in each cluster are respectively read;

B-2: Calculate the feature mean μ of the sample:

Among them, k is the total number of feature i in the electrical cluster, and μ _i is the feature mean of feature i in the cluster;

B-3: Establish the covariance matrix Σ of the characteristic distribution of electricity consumption, X is the eigenvector set of the sample, m is the number of samples,

The multivariate Gaussian distribution model of the user's electricity consumption behavior is obtained.

In this embodiment, the load operation data is used as the input feature in step S2, and the specific calculation steps for calculating the occurrence probability of each cluster corresponding to the operation feature of the electrical appliance by the naive Bayes algorithm are as follows:

C-1: Determine the operating characteristics of the sample set and collect sample data;

C-2: For training samples, the conditional probability of its features is calculated for each category, that is, the probability of occurrence of the corresponding feature value of each feature;

C-3: target classification, read the feature value X of the target, calculate its probability p(X|C _i ) in each category C _i , and take the largest item as the target class.

Figure 2 shows the active power of the fixed-frequency air conditioner-A, electric rice cooker-B, microwave oven-C, induction cooker-D, electric kettle-E and electric water heater-F when they operate independently. The load identification method that invades the load terminal data is used to identify the working conditions of these electrical appliances running alone. The specific identification results are shown in Table 1:

Table 1

Table 1 shows the test results of the load identification method based on the non-intrusive load terminal data on the individual operating conditions of typical electrical appliances. It can be seen that the load identification method can effectively identify electrical appliances.

Claims (7)

1. A load identification method based on non-intrusive load terminal data is characterized in that: the method comprises the following steps:

s1: sampling load data through a non-intrusive load terminal;

s2: the load data is sorted, and the method specifically comprises the following two aspects:

on one hand, the historical electricity utilization data of various electric appliances of the user are subjected to mean shift clustering to obtain different clusters of a clustered electric appliance data set of the user, and load identification characteristics of the electric appliances are extracted from the electric appliance clusters, so that the historical load type load is identified;

on the other hand, a multivariate Gaussian distribution model of the operation characteristics of the electric appliance is established according to the electricity utilization characteristics of the user of each sample point in the cluster, the load operation data is used as input characteristics, and the probability of the operation characteristics of the electric appliance corresponding to each cluster is calculated by a naive Bayes algorithm;

s3: and taking a cluster with the maximum probability as the actual category of the electric appliance, obtaining the electric appliance type of the cluster, and taking the electric appliance type of the cluster as the electric appliance type of the electric appliance.

2. The method of claim 1, wherein the load identification method based on non-intrusive load terminal data is characterized in that: the specific steps of mean shift clustering in step S2 are as follows:

a1: determining the radius r of the sliding window, and starting clustering by using a randomly selected central point C and a circular sliding window with the radius r;

a2: in each iteration process, calculating the mean value in a new sliding window, namely a new central point of the window, wherein the number of points in the sliding window is the density in the window;

a3: continuing to move the sliding window according to the mean value until the density in the circle is not increased any more;

a4: filtering the sliding windows generated in the steps A2 to A4, and when a plurality of windows are overlapped, reserving the window containing the most points and clustering according to the sliding window where the data points are positioned;

a5: and storing the centroid coordinates of each data set as standard characteristic values of the categories, and counting the distribution condition of each characteristic in the group as the distribution range of the electric appliance characteristics.

3. The method of claim 1, wherein the load identification method based on non-intrusive load terminal data is characterized in that: the historical electricity consumption data of the electrical appliance in the step S2 includes total electricity consumption, peak power, start-stop times, start time, and running time of the electrical appliance.

4. The method of claim 2, wherein the load identification method based on non-intrusive load terminal data is characterized in that: the determination process of the radius parameter r of the circular sliding window in step S2 is as follows:

given a dataset P ═ { P (i); 0,1, … n, for any point p (i), calculating the distances between all points in the subset S of the set D from the point p (i) to the point p (i), p (2), …, p (i-1), p (i +1), …, p (n), the distances being sorted in order from small to large, assuming that the sorted distance set is D { D (1), D (2), …, D (k-1), D (k +1), …, D (n) }, D (k) is called k-distance, calculating k-distance for each point p (i) in the set to be clustered, and finally obtaining the k-distance set E of all points E ═ E (1), E (2), …, E (n) }, and according to the obtained k-distance set E of all points, sorting the set E in ascending order to obtain the k-distance set E, obtaining the k-distance set E', and fitting a change curve graph of the k-distance in the sorted E' set, and determining the value of the k-distance corresponding to the position where the change occurs sharply as the value of the radius r according to the change curve graph.

5. The method of claim 1, wherein the load identification method based on non-intrusive load terminal data is characterized in that: and the load identification characteristics of the electrical appliance cluster in the step S2 include load average running time, load average power and running time distribution.

6. The method of claim 1, wherein the load identification method based on non-intrusive load terminal data is characterized in that: the process of establishing the multivariate Gaussian distribution model of the operation characteristics of the electric appliance in the step S2 is as follows:

b-1: data preliminary processing:

respectively reading the user electricity utilization characteristics of the points in each cluster according to different clusters of the clustered user electrical appliance data set;

b-2: calculating the characteristic mean value mu of the sample:

wherein k is the total number of the characteristics i in the electrical appliance cluster, mu_iThe characteristic mean value of the characteristic i in the cluster;

b-3: establishing a covariance matrix sigma of the electrical characteristic distribution, wherein X is a characteristic vector set of the samples, m is the number of the samples,

and obtaining a multivariate Gaussian distribution model of the power utilization behavior of the user.

7. The method of claim 1, wherein the load identification method based on non-intrusive load terminal data is characterized in that: the specific calculation steps of calculating the occurrence probability of each cluster corresponding to the operating characteristics of the electrical appliance by using a naive Bayes algorithm in the step S2 are as follows:

c-1: determining the operation characteristics of a sample set, and collecting sample data;

c-2: training samples, and respectively calculating the conditional probability of the characteristics of each category, namely the occurrence probability of the corresponding characteristic value of each characteristic;

c-3: classifying the target, reading the characteristic value X of the target, calculating it in each class C_iProbability of p (X | C)_i) The class that targets the maximum term.

Images