1. A regularized greedy forest algorithm-based non-intrusive load identification method is characterized by comprising the following steps: the regularization greedy forest algorithm has strong generalization capability when the processing data are unbalanced and the identified tracks with similar characteristics, is low in algorithm operation complexity, improves the identification precision of the algorithm, and specifically comprises the following steps:

step1, preprocessing the data by using the electricity consumption data of residents in a certain area;

step2, selecting a V-I track as a load characteristic, wherein the track characteristic extraction method is to convert an original V-I track into a two-dimensional V-I track through mapping, analyze the correlation between the characteristic track graph and the label of sample data, and use the load characteristic correlated with the label of the sample data to improve the accuracy of load identification;

step3, carrying out load identification by using a non-invasive load identification method based on the regularized greedy forest algorithm, and obtaining an identification result.

2. The method for preprocessing the acquired data and acquiring the related information of the electric equipment of the users in the range as claimed in claim 1, wherein:

the method comprises the following steps that resident electricity consumption data of a certain area are adopted to preprocess the data, and when data imbalance occurs in a sample, the feature training of a few types of samples is insufficient, so that the accuracy deviation occurs in the identification accuracy rate of an algorithm; therefore, the SVMSMOTE algorithm is adopted to expand the few samples so as to generate new data samples; the synthesis strategy is to use a support vector machine classifier to generate support vectors, then generate new minority samples, and finally use the SMOTE oversampling algorithm to synthesize samples.

3. The method of claim 2, wherein the V-I trajectory is selected as a load feature and the correlation of the feature trajectory graph with the label of the sample data is analyzed, wherein:

in a period when the electric equipment stably runs, drawing an original V-I track according to waveform data of high-frequency voltage u and I current, wherein the abscissa is u and the ordinate is I;

the V-I two-dimensional plane is divided into 2N × 2N grids, and the length (i.e., voltage standard value) and height (i.e., current standard value) of each grid are calculated as follows:

initializing a two-dimensional matrix beta with dimension 2 Nx 2N, wherein the elements are all assigned 1, i.e. the color in the grid is initialized to white, for the data point (u) in the original V-I trajectory_j,i_j) (J ═ 1,2, …, J), the index of its position in the matrix β is (x)_j,y_j) If 0 < x_j< 2N +1 and 0 < y_j< 2N +1, the element B (x) of the matrix beta_j,y_j) Set to 0, indicating that the V-I track of the device passes through this cell, marked black:

4. a load recognition using a regularized greedy forest algorithm as claimed in claim 3 wherein:

in a decision tree, such a path from the root node to a child node (whether a leaf node or a non-leaf node) of a sample forms a classification rule, so for a child node v of the decision tree, the rule from the root node to the child node v can be described by the following formula:

function i (x) indicates that the result is 1 when true in parenthesis, and 0 otherwise; if b is_v(x) When x is judged to be 1, the node v can be reached through the judgment of the decision tree node, otherwise, the node v cannot be reached; in the process that a sample goes from a root node to a child node, binary testing (whether the value is greater than a threshold value or less than the threshold value) of a certain characteristic dimension is carried out at each non-leaf node, in the testing of all the passed non-leaf nodes, j tests meet the condition that the value is less than the threshold value of the current non-leaf node, and k tests meet the condition that the value is greater than the threshold value of the current non-leaf node; then each node v (leaf node and non-leaf node) in the decision tree can be used as b_v(x) To represent; thus, in a single decision tree, each node v with two child nodes v1, v2 can be represented as a combination of child nodes:

b_v(x)＝b_v1(x)+b_v2(x). (4)

thus, the decision forest model can be analogized to a combined model of leaf nodes, rather than a combination of decision tree models:

when v is not leaf node, a_v＝0；a_vA weight parameter representing a node; f represents a decision forest; with the representation of the leaf combination model of the decision forest model, the regularized greedy forest algorithm can directly learn the greedy forest by directly utilizing the structure of the forest, and is not only used for generating a single decision tree and adding the decision tree into the decision forest during each iteration.

Background

With the wide popularization of the smart power grid, the resident user side becomes one of important consumption ends, through the sub-item metering and real-time feedback of the power consumption of the user, the resident can be guided to generate reasonable power consumption habits by self, the important effect on relieving the energy crisis is played, and meanwhile, the power grid side is helped to deeply explore the energy-saving potential and the demand response potential of the resident end; the non-invasive load monitoring is an implementation way of power consumption item measurement, and the load identification is one of important components of the non-invasive load monitoring, so that the non-invasive load monitoring has important research significance; the invention discloses a regularized greedy forest algorithm-based non-intrusive load identification method, which can improve load identification precision and has good generalization capability and certain application value.

Disclosure of Invention

The invention mainly aims to provide a regularized greedy forest algorithm-based non-intrusive load identification method.

The method comprises the following steps:

step1, preprocessing the data by using the electricity consumption data of residents in a certain area;

step2, selecting a V-I track as a load characteristic, wherein the track characteristic extraction method is to convert an original V-I track into a two-dimensional V-I track through mapping, analyze the correlation between the characteristic track graph and the label of sample data, and use the load characteristic correlated with the label of the sample data to improve the accuracy of load identification;

step3, carrying out load identification by using a non-invasive load identification method based on the regularized greedy forest algorithm, and obtaining an identification result.

The invention discloses a regularized greedy forest algorithm-based non-intrusive load identification method, which comprises the steps of firstly extracting residential electricity consumption data of a certain area, and preprocessing the data; secondly, selecting a V-I track as a load characteristic, wherein the extraction method of the characteristic is to convert the original V-I track into a two-dimensional V-I track by a binary mapping method, simultaneously analyzing the correlation between a characteristic track graph and a label of sample data, and using the load characteristic which is correlated with the label of the sample data to improve the accuracy of load identification; finally, a non-invasive load identification method based on a regularized greedy forest algorithm is used for identifying the load, and an identification result is obtained; the method can improve the load identification precision, avoids the problems that the load identification algorithm such as deep learning has high operation complexity and can not be used for household embedded equipment, and has good generalization capability and certain application value.

Drawings

In order to make the reader more clearly understand the embodiments of this patent, the following brief description of the drawings in the detailed description of this patent is provided:

FIG. 1 is a flow chart of a regularized greedy forest algorithm implemented by the present invention

FIG. 2 is a structural diagram of a regularized greedy forest algorithm-based non-intrusive load identification method

Detailed Description

The invention mainly aims to provide a regularized greedy forest algorithm-based non-intrusive load identification method.

The method comprises the following steps:

the regularization greedy forest algorithm has strong generalization capability when the processing data are unbalanced and the identified tracks with similar characteristics, is low in algorithm operation complexity, improves the identification precision of the algorithm, and specifically comprises the following steps:

step1, preprocessing the data by using the electricity consumption data of residents in a certain area;

step2, selecting a V-I track as a load characteristic, wherein the track characteristic extraction method is to convert an original V-I track into a two-dimensional V-I track through mapping, analyze the correlation between the characteristic track graph and the label of sample data, and use the load characteristic correlated with the label of the sample data to improve the accuracy of load identification;

step3, carrying out load identification by using a non-invasive load identification method based on the regularized greedy forest algorithm, and obtaining an identification result.

2. The method for preprocessing the acquired data and acquiring the related information of the electric equipment of the users in the range as claimed in claim 1, wherein:

the method comprises the following steps that resident electricity consumption data of a certain area are adopted to preprocess the data, and when data imbalance occurs in a sample, the feature training of a few types of samples is insufficient, so that the accuracy deviation occurs in the identification accuracy rate of an algorithm; therefore, the SVMSMOTE algorithm is adopted to expand the few samples so as to generate new data samples; the synthesis strategy is to use a support vector machine classifier to generate support vectors, then generate new minority samples, and finally use the SMOTE oversampling algorithm to synthesize samples.

3. The method of claim 2, wherein selecting a V-I trajectory as the load feature and analyzing the correlation of the feature trajectory graph with the tags of the sample data comprises:

in a period when the electric equipment stably runs, drawing an original V-I track according to waveform data of high-frequency voltage u and I current, wherein the abscissa is u and the ordinate is I;

the V-I two-dimensional plane is divided into 2N × 2N grids, and the length (i.e., voltage standard value) and height (i.e., current standard value) of each grid are calculated as follows:

initializing a two-dimensional matrix beta with dimension 2 Nx 2N, wherein the elements are all assigned 1, i.e. the color in the grid is initialized to white, for the data point (u) in the original V-I trajectory_j,i_j) (J ═ 1,2, …, J), the index of its position in the matrix β is (x)_j,y_j) If 0 < x_j< 2N +1 and 0 < y_j< 2N +1, the element B (x) of the matrix beta_j,y_j) Set to 0, indicating that the V-I track of the device passes through this cell, marked black:

4. a load recognition using a regularized greedy forest algorithm as claimed in claim 3 wherein:

in a decision tree, such a path from the root node to a child node (whether a leaf node or a non-leaf node) of a sample forms a classification rule, so for a child node v of the decision tree, the rule from the root node to the child node v can be described by the following formula:

function i (x) indicates that the result is 1 when true in parenthesis, and 0 otherwise; if b is_v(x) When x is judged to be 1, the node v can be reached through the judgment of the decision tree node, otherwise, the node v cannot be reached; in the process that a sample goes from a root node to a child node, binary testing (whether the value is greater than a threshold value or less than the threshold value) of a certain characteristic dimension is carried out at each non-leaf node, in the testing of all the passed non-leaf nodes, j tests meet the condition that the value is less than the threshold value of the current non-leaf node, and k tests meet the condition that the value is greater than the threshold value of the current non-leaf node; then each node v (leaf node and non-leaf node) in the decision tree can be used as b_v(x) To represent; thus, in a single decision tree, each node v with two child nodes v1, v2 can be represented as a combination of child nodes:

b_v(x)＝b_v1(x)+b_v2(x). (4)

thus, the decision forest model can be analogized to a combined model of leaf nodes, rather than a combination of decision tree models:

when v is not leaf node, a_v＝0；a_vA weight parameter representing a node; f represents a decision forest; with the representation of the leaf combination model of the decision forest model, the regularized greedy forest algorithm can directly learn the greedy forest by directly utilizing the structure of the forest, and is not only used for generating a single decision tree and adding the decision tree into the decision forest during each iteration.

The patent has certain universality, and various equivalent transformations are carried out on the technical scheme of the invention within the technical idea scope of the invention, and the direct or indirect application in other related technical fields is within the patent protection scope of the invention.