A non-intrusive load decomposition method based on user temporal characteristics and multi-feature GEP

Abstract

The application provides a non-intrusive load decomposition method based on user time characteristics and multi-feature GEP, uses a density-based DBSCAN algorithm, combines time segmentation design power and current characteristic parameters, and uses power and current harmonics as load characteristics. The user daily time is segmented, and the user load use probability in different time periods is calculated. The multi-feature GEP algorithm is used to improve the optimal function, and the time segmentation probability characteristics are combined to accurately decompose and identify the states of various electrical equipment. Compared with the intrusive type, the non-intrusive type only sets monitoring equipment at the total collection end of the user entrance and measures and analyzes the same, and the composition and state of various loads in the system are obtained. From the user-oriented point of view, the result of the non-intrusive load decomposition can not only prompt the user which setting may exist in an abnormal fault, but also help the user to understand the power consumption condition in detail and effectively save energy.

Description

Technical Field

[0001] This invention belongs to the field of power system decomposition technology, and particularly relates to a non-intrusive load decomposition method based on user time characteristics and multi-feature GEP. Background Technology

[0002] With the rapid development of the era of intelligence and information technology, smart grid platforms have begun to take shape. A smart grid, based on modern communication networks, utilizes advanced hardware, efficient sensing, measurement, and communication technologies, and sophisticated control theories and decision-making methods to achieve safe, reliable, economical, efficient, and environmentally friendly power grid operation. Power system load monitoring is a crucial feature of smart grids. However, traditional intrusive load monitoring methods monitor different types of loads separately. This requires a large number of hardware devices, incurring significant costs in procurement, installation, and maintenance. Furthermore, because all testing devices rely on a single data source, data processing and analysis become difficult when the load is excessive.

[0003] Non-intrusive load balancing refers to installing monitoring equipment only at the main data acquisition point at the user's inlet, and then measuring and analyzing it to determine the composition and status of various loads in the system. This saves a lot of monitoring equipment and data, and allows for real-time monitoring of various types of loads. From a user perspective, the results of non-intrusive load balancing not only alert users to potential abnormal faults in their settings, but also help them understand their electricity usage in detail, effectively saving energy. From a national perspective, conducting research on residential electricity consumption issues plays a vital role in energy conservation, emission reduction, and green development of the energy system.

[0004] In view of the current characteristics of a wide variety of power equipment and diverse user features, this invention proposes a non-intrusive load decomposition method based on user time characteristics and multiple features using GEP (Gene Expression Programming). This method comprehensively considers the non-electrical characteristics of the user's appliance usage probability at different time periods, as well as electrical characteristics such as power and current. The time-segmented appliance state probability is used as a new load feature to assist load decomposition. Considering the user's appliance usage time periods, power and current peaks are used as load features. The GEP algorithm is adopted, and the optimal function is improved to accurately decompose and identify the state of various electrical equipment. The GEP-based non-intrusive load decomposition technology mainly considers two aspects: (1) how to extract and establish a feature library based on the user's segmented time probability characteristics, power, and current characteristics; and (2) how to decompose the total load using Gene Expression Programming (GEP). Summary of the Invention

[0005] Technical Problem: The purpose of this invention is to provide a non-invasive load decomposition method based on time segmentation characteristics and gene expression programming. This method enables load decomposition in power systems, allowing for reasonable load decomposition and monitoring, improving data processing efficiency, and ensuring the safe and stable operation of the power system.

[0006] Technical Solution: To address the aforementioned technical problems, this invention proposes a non-intrusive load decomposition method based on user time characteristics and multi-feature GEP. This method includes the following steps:

[0007] Step 1: Collect power and current data from the sample device and process the data;

[0008] Step 2: Perform cluster analysis on the power using the density-based DBSCAN clustering algorithm that can remove noise points, calculate the average value of each cluster, and determine the characteristic parameters of the power steady-state data;

[0009] Step 3: Perform Fast Fourier Transform on the current data to calculate the harmonic values ​​of the current, determine the characteristic parameters of the steady-state current data, and normalize the sampled power, current data and characteristic parameters.

[0010] Step 4: Based on the frequency of device use by users at different times, perform cluster analysis on each appliance, divide the 24 hours of a day into several time periods, and perform probability statistics on the frequency of appliance use by users in each time period to create an appliance probability feature table.

[0011] Step 5: Set the population size, number of generations, selection method (roulette wheel selection), crossover method (single-point crossover), and mutation method (uniform mutation).

[0012] Step 6: Encode the operating status of electrical loads in binary mode. 0 indicates that the appliance is in the off state and 1 indicates that the appliance is in the running state. The number of electrical loads m in the system is equal to the length of the individual. The position of each gene is determined by the node encoding method.

[0013] Step 7: Based on the above encoding method, randomly generate m individuals to form the initial population {x1, x2, x3...x}. m Each individual is an m-dimensional 0-1 vector. For any given time, the total active power sampled is P, and the total reactive power is Q. The active and reactive power characteristics of the m appliances during operation are represented by P. i = [P1, P2, ..., P m Q i = [Q1, Q2, ..., Q m If , then individual i in the population corresponds to a fitted total active power. Fitting total reactive power

[0014] Step 8: Construct the fitness function of the genetic algorithm based on the superposition of current and power;

[0015] Step 9: Calculate the fitness. If the fitness value is satisfied, the process ends and the optimal individual is obtained. Combined with the appliance probability feature table, a secondary selection is performed on the appliance decomposition obtained by the GEP algorithm to confirm the result. Proceed to Step 12; otherwise, proceed to Step 10.

[0016] Step 10: Calculate the objective function value of each generation of individuals in the population through selection, replication, mutation, and gene recombination to obtain the latest population, and retain the best individuals from the previous generation to increase the speed of identifying the best individuals and simplify the iterative convergence process, then proceed to step 11.

[0017] Step 11: Determine if the current iteration count is satisfied. If the iteration count is satisfied, combine the best individual obtained by the final convergence with the appliance probability feature table to perform a secondary selection confirmation on the appliance decomposition obtained by the GEP algorithm, and proceed to Step 12; otherwise, return to Step 9.

[0018] Step 12: If multiple optimal individuals are generated, select the appliance with the highest probability in the current time segment as the unique optimal individual; if only one optimal individual is generated, then adjust it with the appliance probability feature table, that is, if the appliance is not in the probability feature table, then add the appliance to the probability feature table; once the unique optimal individual of the appliance is determined, output the optimal working state sequence to complete the load decomposition.

[0019] Furthermore, the data processing method in step (1) is to filter the data to remove noise.

[0020] Furthermore, in step (2), the active power P and reactive power Q are calculated as follows:

[0021]

[0022]

[0023] Where k is the harmonic order; U k The voltage is the kth harmonic; I k For the kth harmonic current; Φ k It is the phase difference between the voltage and current of the kth harmonic.

[0024] Furthermore, in step (3), the current characteristic parameter is: Among them, H thd G is the current characteristic parameter; k is the harmonic order; G kG1 is the effective value of the kth harmonic component of the current; G2 is the effective value of the fundamental component of the current; and n refers to the total number of harmonics.

[0025] Furthermore, in step (4), the clustering is divided into four time periods: time period 1: 0:00-6:00, time period 2: 6:00-9:00, time period 3: 9:00-17:00, and time period 4: 17:00-24:00.

[0026] Furthermore, the fitness function in step (8) is as follows:

[0027]

[0028] α+β+γ=1

[0029] Among them, P i Q i H represents the active power characteristic and reactive power characteristic of the i-th electrical appliance during operation, respectively; i Let x be the current characteristic of the i-th electrical appliance during operation; i Let P represent the i-th individual in the population, i∈[1,m]; m is the number of electrical appliances; P is the total active power value of the electrical appliances during operation; Q is the total reactive power value of the electrical appliances during operation; H is the total current harmonic value of the electrical appliances during operation; α, β, γ are weighting factors, α, β, γ∈[0,1].

[0030] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects:

[0031] (1) The present invention proposes a non-intrusive load decomposition technology based on GEP improvement. Compared with single power characteristics, the present invention uses multiple features to improve the accuracy of load decomposition. It also combines the probability of users using electrical appliances in different time periods, which can effectively identify various household appliances in different scenarios. Compared with intrusive methods, only the monitoring device is installed at the main end of the power system, saving a lot of monitoring equipment, reducing data processing and analysis work, and indirectly realizing the status monitoring of each type of load in the system.

[0032] (2) Load Feature Extractor: The load feature extractor uses a density-based DBSCAN algorithm to design power feature parameters. The current frequency domain characteristics are analyzed using Fast Fourier Transform to obtain the total harmonic distortion coefficient, and current feature parameters are designed accordingly. Probabilistic analysis is performed on the usage of various appliances by users at different times, and an algorithm that does not require pre-specifying the number of clusters is used to extract probabilistic features for working and resting periods. Each appliance may have one or more states, and different states are likely to have different power values. Clustering can be used to obtain the working state of each appliance. Therefore, the working state of the load can be described by power and current feature parameters.

[0033] (3) Genetic Optimization Load Decomposer: An improved GEP (Genetic Extraction Process) is used to optimize the feature dataset based on combinatorial optimization load decomposers. Power and current features are normalized to construct a fitness function and initial population. Fitness is calculated, and the objective function value of each generation of individuals is calculated through selection, replication, mutation, and gene recombination. The optimal solution is searched step-by-step, and the optimal individual is selected as the output. Because the probability of an appliance switching on and off immediately is very small, retaining individuals from the previous generation increases the speed of identifying the optimal individual and simplifies the iterative convergence process. The final converged optimal individual is combined with the probability distribution map of each appliance's time segment to obtain the final load decomposition. Attached Figure Description

[0034] Figure 1 It is a system architecture diagram, including: load feature extractor and GEP-based load decomposer.

[0035] Figure 2 This is a flowchart illustrating the method of the present invention. Detailed Implementation

[0036] like Figure 2 As shown, this invention proposes a non-intrusive load decomposition method based on user time characteristics and multi-feature GEP, which includes the following steps:

[0037] Step 1: Collect power and current data from the sample device and process the data;

[0038] Step 2: Perform cluster analysis on the power using the density-based DBSCAN clustering algorithm that can remove noise points, calculate the average value of each cluster, and determine the characteristic parameters of the power steady-state data;

[0039] Step 3: Perform Fast Fourier Transform on the current data to calculate the harmonic values ​​of the current, determine the characteristic parameters of the steady-state current data, and normalize the sampled power, current data and characteristic parameters.

[0040] Step 4: Based on the frequency of device use by users at different times, perform cluster analysis on each appliance, divide the 24 hours of a day into several time periods, and perform probability statistics on the frequency of appliance use by users in each time period to create an appliance probability feature table.

[0041] Step 5: Set the population size, number of generations, selection method (roulette wheel selection), crossover method (single-point crossover), and mutation method (uniform mutation).

[0042] Step 6: Encode the operating status of electrical loads in binary mode. 0 indicates that the appliance is in the off state and 1 indicates that the appliance is in the running state. The number of electrical loads m in the system is equal to the length of the individual. The position of each gene is determined by the node encoding method.

[0043] Step 7: Based on the above encoding method, randomly generate m individuals to form the initial population {x1, x2, x3...x}. m Each individual is an m-dimensional 0-1 vector. For any given time, the total active power sampled is P, and the total reactive power is Q. The active and reactive power characteristics of the m appliances during operation are represented by P. i = [P1, P2, ..., P m Q i = [Q1, Q2, ..., Q m If , then individual i in the population corresponds to a fitted total active power. Fitting total reactive power

[0044] Step 8: Construct the fitness function of the genetic algorithm based on the superposition of current and power;

[0045] Step 9: Calculate the fitness. If the fitness value is satisfied, the process ends and the optimal individual is obtained. Combined with the appliance probability feature table, a secondary selection is performed on the appliance decomposition obtained by the GEP algorithm to confirm the result. Proceed to Step 12; otherwise, proceed to Step 10.

[0046] Step 10: Calculate the objective function value of each generation of individuals in the population through selection, replication, mutation, and gene recombination to obtain the latest population, and retain the best individuals from the previous generation to increase the speed of identifying the best individuals and simplify the iterative convergence process, then proceed to step 11.

[0047] Step 11: Determine if the current iteration count is satisfied. If the iteration count is satisfied, combine the best individual obtained by the final convergence with the appliance probability feature table to perform a secondary selection confirmation on the appliance decomposition obtained by the GEP algorithm, and proceed to Step 12; otherwise, return to Step 9.

[0048] Step 12: If multiple optimal individuals are generated, select the appliance with the highest probability in the current time segment as the unique optimal individual; if only one optimal individual is generated, then adjust it with the appliance probability feature table, that is, if the appliance is not in the probability feature table, then add the appliance to the probability feature table; once the unique optimal individual of the appliance is determined, output the optimal working state sequence to complete the load decomposition.

[0049] Furthermore, the data processing method in step (1) is to perform filtering to remove noise.

[0050] Furthermore, in step (2), the active power P and reactive power Q are calculated as follows:

[0051]

[0052]

[0053] Where k is the harmonic order; U k The voltage is the kth harmonic; I k For the kth harmonic current; Φ k It is the phase difference between the voltage and current of the kth harmonic.

[0054] Furthermore, in step (3), the current characteristic parameter is: Among them, H thd G is the current characteristic parameter; k is the harmonic order; G k G1 is the effective value of the kth harmonic component of the current; G2 is the effective value of the fundamental component of the current; and n refers to the total number of harmonics.

[0055] Furthermore, in step (4), the clustering is divided into four time periods: time period 1: 0:00-6:00, time period 2: 6:00-9:00, time period 3: 9:00-17:00, and time period 4: 17:00-24:00.

[0056] Furthermore, the fitness function in step (8) is as follows:

[0057]

[0058] α+β+γ=1

[0059] Among them, P i Q i H represents the active power characteristic and reactive power characteristic of the i-th electrical appliance during operation, respectively; i Let x be the current characteristic of the i-th electrical appliance during operation; i Let P represent the i-th individual in the population, i∈[1,m]; m is the number of electrical appliances; P is the total active power value of the electrical appliances during operation; Q is the total reactive power value of the electrical appliances during operation; H is the total current harmonic value of the electrical appliances during operation; α, β, γ are weighting factors, a, β, γ∈[0,1].

Claims

1. A non-intrusive load decomposition method based on user temporal characteristics and multi-feature GEP, characterized in that, The method comprises the following steps: Step 1: Collecting power and current data of the sample equipment, and performing data processing; Step 2: Clustering analysis of power is performed according to a density-based DBSCAN clustering algorithm capable of removing noise points, the average value of each class is calculated, and the characteristic parameters of the power steady-state data are determined; Step 3: Fast Fourier transform processing is performed on the current data, the harmonic values of the current are calculated, the characteristic parameters of the current steady-state data are determined, and the normalized processing is performed on the sampled power and current data and the characteristic parameters; Step 4: According to the frequency of equipment use of the user in different time periods, clustering analysis is performed on each electrical appliance, 24 hours of a day is divided into several time periods, the use frequency of the user on the electrical appliance in each time period is statistically analyzed, and an electrical appliance probability feature table is prepared; Step 5: The population size of GEP is set, the number of generations is set, the roulette method is selected, the single-point crossover is selected, and the uniform mutation is selected; Step 6: The running state of the electrical appliance load is encoded in a binary manner, 0 represents that the electrical appliance is in the off state, 1 represents that the electrical appliance is in the running state, the number m of electrical appliance loads in the system is equal to the length of the individual, and each gene position is determined by the node coding mode; Step 7: According to the coding mode described above, m individuals are randomly generated to constitute an initial population {x1, x2, x3, …, xm}, each individual being an m-dimensional 0-1 vector, for any total active power P and total reactive power Q sampled at any time, the active and reactive power characteristics of the m electrical appliances in operation are represented as m , , , then the population individual i corresponds to a fitted total active power and a fitted total reactive power ; Step 8: The fitness function of the genetic algorithm is constructed according to the current superposition and the power superposition; Step 9: The fitness is calculated, if the fitness value is satisfied, the optimal individual is obtained, the electrical appliance decomposition obtained by the GEP algorithm is confirmed again according to the electrical appliance probability feature table, and step 12 is entered, otherwise step 10 is entered; Step 10: The target function value in each generation population individual is calculated through selection, replication, mutation and gene recombination, the latest population is obtained, the optimal individual of the last generation is kept to increase the identification speed of the optimal individual and simplify the iteration convergence process, and step 11 is entered; Step 11: Whether the current iteration number is satisfied is judged, if the iteration number is satisfied, the optimal individual obtained through the last convergence is combined with the electrical appliance probability feature table, the electrical appliance decomposition obtained by the GEP algorithm is confirmed again, and step 12 is entered; otherwise, the step 9 is returned; Step 12: If multiple optimal individuals are generated, the electrical appliance with the highest use probability in the current time period is selected as the only optimal individual; if there is only one optimal individual, the electrical appliance probability feature table is corrected, that is, if there is no such electrical appliance in the probability feature table, the electrical appliance is added to the probability feature table; after the only electrical appliance optimal individual is determined, the optimal working state sequence is output, and the load decomposition is completed; The fitness function in step 8 is as follows: ； wherein, , are the active and reactive power signatures of the i-th appliance when it is running, respectively; is the current signature when the i-th appliance is running; denotes the i-th individual in the population, ; is the number of appliances; is the total active power value when the appliances are running; is the total reactive power value when the appliances are running; is the total current harmonic value when the appliances are running, is a weight factor, ∈ [0, 1].

2. The non-intrusive load disaggregation method based on user temporal characteristics and multi-feature GEP of claim 1, wherein, The data processing method in step 1 is that the data is filtered to remove noise. 3.The non-intrusive load disaggregation method based on user temporal characteristics and multi-feature GEP of claim 1, wherein, In step 2, the active power P and the reactive power Q are calculated as follows: ； ； wherein, is the harmonic number; is the sub-harmonic voltage; is the sub-harmonic current; is the phase difference of the voltage and current of the sub-harmonic.

4. The non-intrusive load disaggregation method based on user temporal characteristics and multi-feature GEP of claim 1, wherein, In step 3, the characteristic parameter is wherein is the current characteristic parameter; is the harmonic number; is the current effective value of the nth harmonic component; is the effective value of the fundamental component of the current, and n is the total number of harmonics.

5. The non-intrusive load disaggregation method based on user temporal characteristics and multi-feature GEP of claim 1, wherein, In step 4, four time periods are divided, time period one: 0:00-6:00, time period two: 6:00-9:00, time period three: 9:00-17:00, and time period four: 17:00-24:00.

Images