A non-intrusive load decomposition method based on waveform distortion feature time sequence change

Abstract

The present application relates to the technical field of non-intrusive load decomposition, in particular to a non-intrusive load decomposition method based on waveform distortion feature time sequence change, comprising the following steps: real-time acquisition of current signal of a to-be-monitored power circuit in units of 1 second; equal-length cutting of the acquired current signal in units of complete current cycles to form a current signal subsequence; calculation of the effective value and waveform distortion feature of the current signal subsequence to obtain a current signal state feature matrix; beneficial effects are that: the mean value of the remaining elements of the new current signal feature matrix is processed to obtain a power load feature vector, and load decomposition of the to-be-monitored power network is realized according to the effective power load feature vector time sequence change rule; compared with traditional non-intrusive load decomposition technology, the technical problems of insufficient feature migration, small load state feature difference, and poor dynamic working condition adaptability can be effectively solved, and the method is more practical due to the small demand for computing resources.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of non-intrusive load decomposition, in particular to a non-intrusive load decomposition method based on waveform distortion feature time sequence change. BACKGROUND

[0002] Electric load decomposition is an important research direction in the field of intelligent electricity and safe electricity, and is of great significance to energy management and healthy state monitoring of electricity network. With the gradual improvement of users' demand for fine management of electricity and positioning of electricity safety hazards, it is urgent to obtain a technical means to analyze the specific working state of various loads in the monitored electricity network, and provide prior topological constraints for the rapid positioning of electrical safety hazards when necessary.

[0003] Based on the above situation, non-intrusive load decomposition technology has become an important means for precise monitoring of load state due to its easy installation and low cost, but on the one hand, the frequent operation of high-power electrical appliances will mask the state characteristics of low-power electrical appliances, and on the other hand, the diversity of electrical loads and the similarity of electrical parameters of electrical loads make it difficult for existing non-intrusive load monitoring equipment to adapt to different electricity scenes.

[0004] Therefore, there is an urgent need for a new framework of non-intrusive load decomposition technology that takes into account the adaptability of low-power electrical appliances, dynamic changes of electrical equipment, efficiency of edge computing, and sufficient distinction of load characteristics, in order to meet the fine needs of intelligent power grid energy efficiency management and load state monitoring. SUMMARY

[0005] The purpose of the present application is to provide a non-intrusive load decomposition method based on waveform distortion feature time sequence change, to solve the technical problems of insufficient feature transferability, small load state feature difference and poor dynamic working condition adaptability of existing non-intrusive load decomposition technology in engineering application, and to realize precise monitoring of energy efficiency management and load state of the monitored electricity network.

[0006] To achieve the above purpose, the present application provides the following technical scheme: a non-intrusive load decomposition method based on waveform distortion feature time sequence change, comprising the following steps:

[0007] The current signal of the monitored electricity circuit is converted into a voltage signal recognizable by the embedded system by using a high-precision current transformer, and a synchronous sampling and real-time flow processing architecture is adopted to collect the current signal of the monitored electricity circuit in real time in units of 1 second;

[0008] The collected current signal is cut into equal length in units of complete current period to form the current signal subsequence;

[0009] The effective value, maximum value and a difference between the effective value and a value of the third harmonic included in the current signal subsequence, a time domain feature vector of the current signal subsequence, and a frequency domain feature vector of the current signal subsequence;

[0010] determining a local maximum value of the current signal subsequence difference signal and a maximum value near the balance position, and forming a trend change feature of the current signal;

[0011] combining the time domain feature vector, the frequency domain feature vector, and the trend change feature vector of the current signal subsequence in a certain order to form a current signal state feature matrix;

[0012] eliminating accidental error elements in the current signal state feature matrix based on an outlier index of the trend change feature vector of the current signal state feature matrix to form a new current signal state feature matrix;

[0013] performing mean value processing on the remaining elements of the new current signal feature matrix to obtain an electrical load feature vector;

[0014] judging whether the electrical load feature vector is a valid feature vector according to a discrete scale of the effective value column of the current signal state feature matrix;

[0015] based on the real-time and continuity of data acquisition and processing, obtaining a time sequence change matrix of the valid electrical load feature vector within a certain time window;

[0016] performing classification processing on electrical loads in a to-be-monitored electrical circuit according to the time sequence change characteristics of the valid electrical load feature vector, and realizing load decomposition of the to-be-monitored electrical circuit.

[0017] Preferably, the current signal acquisition of the to-be-monitored electrical circuit is a signal acquisition and processing parallel operation scheme, and the signal acquisition and processing parallel operation scheme comprises:

[0018] using a synchronous sampling and real-time flow processing architecture, performing acquisition-processing pipeline through a ring buffer driven by DMA to ensure the integrity and gapless continuity of the full-link current signal.

[0019] Preferably, the waveform distortion feature of the current signal subsequence includes a local maximum value of the current signal subsequence difference signal, a maximum value near the balance position, a difference between the effective value and a value of the third harmonic included in the current signal subsequence, and an amplitude of the third harmonic included in the current signal subsequence.

[0020] calculating the effective value of the current signal subsequence, ​​,

[0021]

[0022] determining the maximum value of the subsequence of the current signal and its index value ;

[0023] calculating the difference sequence of , ,

[0024]

[0025] determining the local maximum value of the difference sequence difference signal and the maximum value near the balance position, specifically determining the maximum value in the interval , and the maximum value in the interval , ;

[0026] determining the difference between the maximum value of the subsequence of the current signal and times the effective value ,

[0027]

[0028] determining the amplitude of the third harmonic contained in the subsequence of the current signal , and the specific steps are as follows:

[0029] calculating the sine component of the third harmonic contained in the subsequence of the current signal ,

[0030]

[0031] calculating the cosine component of the third harmonic contained in the subsequence of the current signal ,

[0032]

[0033] calculating the amplitude of the third harmonic contained in the subsequence of the current signal ,

[0034]

[0035] wherein is the nth value of the nth subsequence of the current signal is the maximum value of the nth subsequence of the current signal is the maximum value of the nth subsequence of the current signal is the maximum value of the nth subsequence of the current signal is the maximum value of the nth subsequence of the current signal is the maximum value of the nth subsequence of the current signal​ The position corresponding to the maximum value of each current signal subsequence. is the length of the current signal subsequence.

[0036] Preferably, the current signal state feature matrix is ​​obtained based on the effective value and waveform distortion characteristics of the current signal subsequence. :

[0037] .

[0038] Preferably, outliers in the current signal state feature matrix are removed using the waveform distortion feature column as a reference, and the remaining elements of the current signal feature matrix are averaged to obtain the electrical load feature vector. Specifically, the waveform distortion feature... Using column vectors as a reference, determine The index of outlier points in the column vector is used to remove the current signal state feature matrix based on the index of the outlier points. The elements of the corresponding rows form a new current signal state feature matrix. Furthermore, the current signal state feature matrix, The average value of each column is used to form the characteristic vector of electricity load. The outlier is an element that satisfies the following condition:

[0039]

[0040] Here, for The mean of a vector. for The standard deviation of a vector.

[0041] Preferably, the discrete scale of the effective value column of the current signal state feature matrix is ​​used as the standard to determine whether the electrical load feature vector is a valid feature vector. Specifically, if The following conditions must be met to determine The effective electrical load characteristic vector:

[0042]

[0043] in, express standard deviation The threshold is preset in advance.

[0044] Preferably, the effective electrical load characteristic vector is determined. Subsequently, due to the real-time and continuous nature of data acquisition and processing, data can be generated within a certain time window. The data flow matrix, will the The data stream matrix of the current signal is input into the trained LSTM classification model to realize the load decomposition of the to-be-monitored power circuit.

[0045] Compared with the prior art, the present application has the following advantages:

[0046] The non-intrusive load decomposition method based on waveform distortion feature time sequence change provided by the present application realizes the equal-period truncation of the trunk current signal of the to-be-monitored power network to form a current signal subsequence, and obtains the current signal state feature matrix by calculating the effective value and the waveform distortion feature of the current signal subsequence. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 The method flowchart of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical scheme of the present application clear, complete and the advantages more clear and obvious, the embodiments of the present application are further described in detail below in combination with the drawings. It should be understood that the specific embodiments described here are part of the embodiments of the present application, not all the embodiments, and are only used to explain the embodiments of the present application, and do not limit the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0049] The present application provides a technical scheme: a non-intrusive load decomposition method based on waveform distortion feature time sequence change, comprising the following steps:

[0050] Step 1: The current signal of the to-be-monitored power network is converted into a voltage signal recognizable by the embedded system by a sampling resistor in a certain proportion by a high-precision current transformer, and the voltage signal across the sampling resistor is collected in real time to form a digital sampling sequence in units of 1 second by using synchronous sampling and real-time stream processing architecture, and the current signal is inversely transformed in a certain proportion ;

[0051] Step 2: The current signal is truncated by equal length of period to form a current signal subsequence ;

[0052] It should be noted that the sampling frequency of the current signal in the present embodiment is 40k, and thus the length of the current signal is 40000. Since the power supply for Chinese residents is 50Hz, 220V alternating current, the current signal is truncated into 50 parts of equal length, i.e. ;

[0053] Step 3: Calculate the effective value of the current signal subsequence, and obtain the vector;

[0054] Step 4: Determine the maximum value of the current signal subsequence and its index value , calculate the difference between the maximum value of the current signal subsequence and times the effective value, and obtain the vector;

[0055] Step 5: Calculate the amplitude of the third harmonic contained in the current signal subsequence, and obtain the frequency domain feature vector of the current signal subsequence;

[0056] Step 6: Perform difference operation on the current signal subsequence to determine the difference signal of the current signal subsequence, and determine the maximum value of the difference signal in the interval , and the maximum value of the difference signal in the interval , to form the trend change feature vector of the current signal subsequence;

[0057] Step 7: Combine the above feature vectors in a certain order to form the current signal state feature matrix ;

[0058] Step 8: Determine the index of the outlier in the vector, remove the , , , vector and the elements corresponding to the row of the vector itself based on the index of the outlier, and eliminate accidental errors caused by signal acquisition;

[0059] It should be noted that the accidental errors caused by the signal acquisition process are mainly in the form of burr points on the current signal waveform, which can generally be eliminated by a noise reduction algorithm. In the present embodiment, in order to reduce the consumption of computing resources, the calculation of the waveform distortion feature outliers of the current signal subsequence is used to complete the point elimination of accidental errors, and the interference of the power consumption environment is further eliminated through the mean value calculation of step 9, which can effectively improve the anti-interference of load decomposition.

[0060] Step 9: average the remaining elements of the current signal state feature matrix by column to obtain the electrical load feature vector and the discrete scale of the effective value column of the current signal state feature matrix as the standard to judge whether the electrical load feature vector is an effective feature vector;

[0061] Step 10: based on the real-time and continuity of data acquisition and processing, obtain the time sequence change matrix of the effective electrical load feature vector within a certain time window;

[0062] Step 11: classify the electrical load in the to-be-monitored electrical circuit according to the time sequence change characteristics of the effective electrical load feature vector, and realize the load decomposition of the to-be-monitored electrical circuit.

[0063] Although the embodiments of the present application have been shown and described, it can be understood by those of ordinary skill in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A non-intrusive load decomposition method based on the temporal variation of waveform distortion characteristics, characterized in that: Includes the following steps: The current signal of the power circuit to be monitored is collected in real time in 1-second increments. The acquired current signal is divided into equal-length segments based on a complete current cycle to form the current signal subsequence. Calculate the effective value and waveform distortion characteristics of the current signal subsequence to obtain the current signal state feature matrix; To obtain the difference sequence of the current signal subsequence, determine the local maximum value and the maximum value near the equilibrium position of the difference signal in the difference sequence. Specifically, this involves determining... In the interval The maximum value within, and in the interval The maximum value within, ; Based on the effective values ​​and waveform distortion characteristics of the current signal subsequence, the current signal state feature matrix is ​​obtained. : ; For current signal subsequence The effective value; The amplitude of the third harmonic contained in the current signal subsequence; This is the difference between the maximum value and the effective value of the current signal subsequence; Using the waveform distortion feature column of the current signal state feature matrix as a reference, outliers in the current signal state feature matrix are removed. The remaining elements of the current signal feature matrix are then averaged to obtain the electrical load feature vector. Specifically, the waveform distortion feature... Using column vectors as a reference, determine The index of outlier points in the column vector is used to remove the current signal state feature matrix based on the index of the outlier points. The elements of the corresponding rows form a new current signal state feature matrix. Furthermore, the current signal state feature matrix, The average value of each column is used to form the characteristic vector of electricity load. The outlier is an element that satisfies the following condition: Here, for The mean of a vector. for The standard deviation of a vector; The discrete scale of the effective value column of the current signal state feature matrix is ​​used as the standard to determine whether the power load feature vector is an effective feature vector; Based on the time-series variation characteristics of the effective feature vector of the electrical load, the electrical load in the monitored electrical circuit is classified and processed to achieve load decomposition of the monitored electrical circuit.

2. The non-intrusive load decomposition method based on the temporal variation of waveform distortion characteristics according to claim 1, characterized in that: The current signal acquisition and processing of the monitored power circuit is carried out in parallel using the following scheme: Employing a synchronous sample-and-hold and real-time stream processing architecture, the current signal after anti-aliasing filtering is converted by an ADC, and a DMA-driven ring buffer is used to realize the acquisition-processing pipeline, ensuring the integrity and seamless continuity of the current signal throughout the entire link.

3. The non-intrusive load decomposition method based on the temporal variation of waveform distortion characteristics according to claim 1, characterized in that: The waveform distortion characteristics of the current signal subsequence include the local maximum value of the differential signal of the current signal subsequence, the maximum value near the equilibrium position, and the maximum value of the current signal subsequence. The difference between the effective values ​​and the amplitude of the third harmonic contained in the current signal subsequence; specifically, the steps for calculating the effective value and waveform distortion characteristics of the current signal subsequence include: Calculate the current signal subsequence The effective value, , Determine the current signal subsequence maximum value and its index value ; Seeking Difference sequences , Determine the maximum value of the current signal subsequence and The difference between the effective values ​​and the effective values. , Determine the amplitude of the third harmonic contained in the current signal subsequence. The specific steps are as follows: Calculate the sinusoidal component of the third harmonic contained in the current signal subsequence. , Calculate the cosine component of the third harmonic contained in the current signal subsequence. , Calculate the amplitude of the third harmonic contained in the current signal subsequence. , in, For the first The first current signal subsequence One value, For the first The maximum value of each current signal subsequence. For the first The position corresponding to the maximum value of each current signal subsequence. is the length of the current signal subsequence.

4. The non-intrusive load decomposition method based on the temporal variation of waveform distortion characteristics according to claim 1, characterized in that: The discrete scale of the effective value column of the current signal state feature matrix is ​​used as the standard to determine whether the electrical load feature vector is an effective feature vector. Specifically, if The following conditions must be met to determine The effective electrical load characteristic vector: in, express standard deviation The threshold is preset in advance.

5. The non-intrusive load decomposition method based on the temporal variation of waveform distortion characteristics according to claim 1, characterized in that: Determine the characteristic vector of effective electrical load. Subsequently, due to the real-time and continuous nature of data acquisition and processing, data can be generated within a certain time window. The data flow matrix, will the By inputting the data flow matrix into the trained LSTM classification model, the electrical load decomposition of the power circuit to be monitored can be achieved.

Images