Load distinguishing method and device

Abstract

The invention discloses a load discrimination method and device, and relates to the electrical field, and the method comprises the steps: collecting the electrical data of a user total ammeter side, and the electrical data comprise voltage data and current data; performing feature extraction on the electrical data, wherein features comprise at least one of power class features, harmonic and electric energy quality features, instantaneous features and statistical distribution features; performing feature fusion on the features to obtain fusion data; and inputting the fusion data into a pre-trained neural network, so that the neural network outputs a load type corresponding to the electrical data, and the load type comprises an adjustable load and a non-adjustable load. According to the method, the electrical data is represented through the multiple feature data, the type of the load is determined based on the fusion data obtained after fusion of the multiple features, and the type of the load is distinguished on the premise of ensuring user privacy, so that subsequent power grid adjustment is facilitated.

Description

TECHNICAL FIELD

[0001] The present application relates to the electrical field, in particular to a load discrimination method and device. BACKGROUND

[0002] With the development of smart grid and demand response, user-side load management has become an important part of fine peak regulation of power systems. Especially, accurately distinguishing between adjustable loads (such as air conditioners and water heaters) and non-adjustable loads (such as lighting and communication equipment) is of great significance to ensure continuous power supply for critical loads. The non-intrusive load identification method in the related art mainly relies on time domain or frequency domain signal features, which has the following disadvantages: lack of unified electrical metering feature definition, inability to fully quantify load behavior, difficulty in simultaneously considering transient start / stop information and steady-state operating characteristics, insufficient classification accuracy and interpretability, and limited engineering application. SUMMARY

[0003] The purpose of the present application is to provide a load discrimination method and device, which characterizes electrical data through multiple feature data, determines the type of load based on fusion data after fusion of multiple features, and discriminates the type of load under the premise of ensuring user privacy, so as to facilitate subsequent power grid regulation.

[0004] To solve the above technical problems, the present application provides a load discrimination method, comprising:

[0005] Collecting electrical data on the user's total electric meter side, the electrical data including voltage data and current data;

[0006] Extracting features from the electrical data, the features including at least one of power features, harmonic and power quality features, instantaneous features and statistical distribution features;

[0007] Fusing the features to obtain fusion data;

[0008] Inputting the fusion data into a pre-trained neural network, so that the neural network outputs the load type corresponding to the electrical data, the load type including adjustable load and non-adjustable load.

[0009] On the other hand, after collecting the electrical data on the user's total electric meter side, it further comprises:

[0010] Using a band-pass filter to filter out the fundamental interference and high-frequency noise in the electrical data;

[0011] Fourier transforming the voltage data and the current data respectively to convert the voltage data and the current data from time domain data to frequency domain data;

[0012] Determining the phase angle of the converted voltage data and current data;

[0013] Determine the phase difference between the voltage data and the current data, and the expression for the phase difference is: ;

[0014] in, Let θ be the phase difference. i Let θ be the phase angle of the current data. u The phase angle of the voltage data;

[0015] The phase difference is used to correct the waveform of the current data so that the waveform of the current data is aligned with the waveform of the voltage data.

[0016] On the other hand, after collecting electrical data from the user's main electricity meter, it also includes:

[0017] The electrical data is segmented according to a time window of a first preset length to obtain multiple segments of electrical segmentation data;

[0018] Or determine the trigger point according to preset trigger conditions;

[0019] The electrical data is segmented using a time window of a second preset length where the trigger point is located, resulting in multiple segments of electrical segmentation data.

[0020] Feature extraction of the electrical data includes:

[0021] Feature extraction is performed on each segment of electrical data.

[0022] On the other hand, feature extraction is performed on the electrical data, including:

[0023] The active power, reactive power, and power factor of the electrical data are obtained respectively.

[0024] The expression for the active power is: ;

[0025] The expression for the reactive power is: ;

[0026] The expression for the power factor is: ;

[0027] Where P is the active power, and T is the period of a fundamental frequency. Here is the voltage data at time t. Here, S is the current data at time t, S is the apparent power, the product of the voltage data and the current data, Q is the reactive power, and PF is the power factor;

[0028] The active power, the reactive power and the power factor are taken as power class feature data of the electrical data.

[0029] On the other hand, feature extraction is performed on the electrical data, including:

[0030] A total harmonic distortion rate of the electrical data is determined, and an expression of the total harmonic distortion rate is ;

[0031] A voltage deviation rate of the electrical data is determined, and an expression of the voltage deviation rate is ;

[0032] Wherein, THD is the total harmonic distortion rate, representing the influence degree of the electrical data by harmonics, h is the harmonic order, H is the highest analysis harmonic order, I h is the effective value of the hth harmonic current, I1 is the effective value of the fundamental current, is the voltage deviation rate, representing the deviation degree of the actual operating voltage from the rated voltage, is the actual voltage effective value, is the rated voltage value;

[0033] The total harmonic distortion rate and the voltage deviation rate are taken as harmonic and power quality feature data of the electrical data.

[0034] On the other hand, feature extraction is performed on the electrical data, including:

[0035] A current change rate when the load is started is obtained, and an expression of the starting current change rate is ;

[0036] A current change rate when the load is turned off is obtained, and an expression of the off current change rate is ;

[0037] Total energy consumed or released by the load in the start-stop process is obtained, and an expression of the total energy is ;

[0038] Wherein, is the starting current change rate, is the off current change rate, is the current change amount, is the time window width, is the transient process energy, is voltage data at time t, is current data at time t, is the transient starting time;

[0039] The starting current change rate, the closing current change rate and the total energy are taken as instantaneous feature data of the electrical data.

[0040] On the other hand, feature extraction is performed on the electrical data, including:

[0041] Obtaining skewness of the electrical data, the expression of the skewness is ;

[0042] Obtaining kurtosis of the electrical data, the expression of the kurtosis is ;

[0043] Wherein, the skewness is the skewness, representing the asymmetry of the waveform distribution of the electrical data, the kurtosis is the kurtosis, representing the sensitivity to extreme values, x is the sequence of the electrical data, μ is the mean of the sequence of the electrical data, is the standard deviation of the sequence of the electrical data;

[0044] The skewness and the kurtosis are taken as statistical distribution features of the electrical data.

[0045] On the other hand, before the feature fusion of the features is performed to obtain fusion data, further including:

[0046] Normalizing the feature data of each feature;

[0047] Determining the dimension of the normalized feature data;

[0048] Extending the dimensions of all feature data to the same dimension as the feature data with the largest dimension;

[0049] Splicing the feature data after dimension extension to obtain fusion data.

[0050] On the other hand, the feature fusion of the features is performed to obtain fusion data, including:

[0051] Dividing a weight for each feature data;

[0052] Multiplying each feature data with the corresponding weight, and accumulating the products of all feature data and the corresponding weights to obtain fusion data;

[0053] Inputting the fusion data into a pre-trained neural network, so that the neural network outputs a load type corresponding to the electrical data, including:

[0054] Inputting the fusion data into a pre-trained neural network, so that the neural network outputs a discriminant value;

[0055] Determining the load type based on the discriminant value.

[0056] To solve the above technical problems, the application further provides a load discrimination device, comprising:

[0057] a memory for storing a computer program;

[0058] a processor for executing the computer program to realize the steps of the load discrimination method.

[0059] The application provides a load discrimination method and device, relates to the electrical field, and comprises the following steps: collecting electrical data on the side of a user's total electric meter, wherein the electrical data comprises voltage data and current data; performing feature extraction on the electrical data, wherein the features comprise at least one of power features, harmonic and power quality features, instantaneous features and statistical distribution features; performing feature fusion on the features to obtain fusion data; and inputting the fusion data into a pre-trained neural network so that the neural network outputs a load type corresponding to the electrical data, wherein the load type comprises adjustable load and non-adjustable load. The electrical data is represented by various feature data, the type of the load is determined based on the fusion data obtained by fusing the various features, the type of the load is discriminated under the premise of ensuring user privacy, and subsequent power grid regulation is facilitated. BRIEF DESCRIPTION OF DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required to be used in the prior art and the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0061] Figure 1 A flowchart of a load discrimination method provided by the present application;

[0062] Figure 2 A structural schematic diagram of a load discrimination device provided by the present application. DETAILED DESCRIPTION

[0063] The core of the present application is to provide a load discrimination method and device, which represents electrical data by various feature data, determines the type of the load based on fusion data obtained by fusing the various features, discriminates the type of the load under the premise of ensuring user privacy, and facilitates subsequent power grid regulation.

[0064] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0065] Figure 1 A flowchart of a load discrimination method provided by the present application is shown in the figure. The load discrimination method comprises the following steps.

[0066] S11: Collect electrical data on the side of a total electric meter, the electrical data comprising voltage data and current data;

[0067] With the development of smart grid and demand response, user-side load management has become an important part of fine peak regulation of the power system. In particular, accurately distinguishing between adjustable loads (such as air conditioners and water heaters) and non-adjustable loads (such as lighting and communication equipment) is of great significance to ensuring continuous power supply for critical loads.

[0068] First, the voltage waveform, current waveform and apparent power are synchronously collected on the side of the total electric meter. The sampling rate is , and the sampling period is Then, the collected data is preprocessed.

[0069] S12: Extracting features from the electrical data, the features including at least one of power features, harmonic and power quality features, instantaneous features and statistical distribution features;

[0070] The electrical metering features mainly include power features, harmonic and power quality indicators, instantaneous features and statistical distribution features, mainly but not limited to the above features. By extracting the features, different types of loads can be distinguished more accurately. Since different types of loads have their own unique startup characteristics, the features need to be determined after extraction.

[0071] S13: Feature fusion of the features to obtain fused data;

[0072] Since the obtained data is a feature in multiple directions, the multiple features need to be fused together, and then the calculation is performed according to the single fused data. The calculation process after fusion also reduces the workload.

[0073] During the fusion process, weights can be added to each data according to actual needs. The fused data obtained in this way can more accurately express the features.

[0074] S14: input the fusion data into the pre-trained neural network, so that the neural network outputs the load type corresponding to the electrical data, and the load type includes adjustable load and non-adjustable load.

[0075] Using the neural network model to distinguish the fusion data can more conveniently and quickly obtain the load type corresponding to each fusion data. Multi-domain feature splicing: splicing power class features, harmonics and power quality indicators, instantaneous features and statistical distribution features according to a unified format. Each domain is spliced in a predetermined order, and the same dimension and index mapping are maintained at any time.

[0076] The application provides a load discrimination method, relating to the electrical field, including collecting electrical data on the user's total electric meter side, the electrical data including voltage data and current data; performing feature extraction on the electrical data, the features including at least one of power class features, harmonic and power quality features, instantaneous features and statistical distribution features; performing feature fusion on the features to obtain fusion data; inputting the fusion data into a pre-trained neural network, so that the neural network outputs the load type corresponding to the electrical data, and the load type includes adjustable load and non-adjustable load. The electrical data is characterized by various feature data, and the type of the load is determined based on the fusion data after the fusion of the various features, so as to discriminate the type of the load under the premise of guaranteeing user privacy, so as to facilitate subsequent power grid regulation.

[0077] On the basis of the above embodiments:

[0078] In some embodiments, after collecting the electrical data on the user's total electric meter side, the method further includes:

[0079] Using a band-pass filter to filter out the fundamental interference and high-frequency noise in the electrical data;

[0080] Performing Fourier transform on the voltage data and the current data respectively to convert the voltage data and the current data from time domain data to frequency domain data;

[0081] Determining the phase angle of the converted voltage data and current data;

[0082] Determining the phase difference between the voltage data and the current data, and the expression of the phase difference is ;

[0083] wherein, is the phase difference, θ i is the phase angle of the current data, θ u is the phase angle of the voltage data;

[0084] Using the phase difference to correct the waveform of the current data, so that the waveform of the current data is aligned with the waveform of the voltage data.

[0085] A band-pass filter H(f) is used to remove the power fundamental frequency interference and high frequency switching noise.

[0086] Phase correction: correct the sampling delay deviation of voltage / current;

[0087] Do FFT on voltage and current respectively to extract the fundamental component;

[0088] Calculate the phase difference ;

[0089] Use Correct the current waveform to align with the voltage.

[0090] In some embodiments, after collecting the electrical data on the user's total meter side, further comprising:

[0091] According to the first preset length of the time window, the electrical data is divided to obtain a plurality of electrical division data;

[0092] Or according to the preset trigger condition to determine the trigger point;

[0093] The electrical data is divided according to the second preset length of the time window where the trigger point is located to obtain a plurality of electrical division data;

[0094] Feature extraction is performed on the electrical data, including:

[0095] Feature extraction is performed on each piece of electrical division data respectively.

[0096] Data slicing: a fixed time window (such as 200ms) and an event trigger window (detecting load change points) are combined to divide the signal.

[0097] Explanation of the fixed time window: 2000 points of data are intercepted every 200ms.

[0098] Explanation of the event trigger window: when the current moment rises (such as the start of the air conditioner compressor), the trigger point is taken as the center, and 50ms before + 150ms after, a total of 200ms, are intercepted.

[0099] In some embodiments, the feature extraction is performed on the electrical data, including:

[0100] Active power, reactive power and power factor of the electrical data are obtained respectively;

[0101] The expression of active power is ;

[0102] The expression of reactive power is ;

[0103] The expression of power factor is ;

[0104] wherein P is the active power, T is the period of a fundamental wave, is the voltage data at time t, is the current data at time t, S is the apparent power, the product of the voltage data and the current data, Q is the reactive power, and PF is the power factor;

[0105] The active power, the reactive power, and the power factor are taken as power class characteristic data of the electrical data.

[0106] The active power is the part of the circuit that is actually consumed and converted into other forms of energy, and directly corresponds to the actual output capacity of the equipment, such as light bulb light emission, motor driving load, and resistance heating. The active power is the core basis for electricity billing because it is the energy actually consumed by the user.

[0107] The reactive power does not directly consume electric energy, but performs periodic energy exchange between the power supply and the load (such as inductance and capacitance). It is mainly used to establish and maintain the magnetic field or electric field required for equipment operation, such as transformers and motors that require reactive power to generate a magnetic field to work. The reactive power does not produce actual utility, but occupies the transmission capacity of the power grid, resulting in increased line loss. The power factor is the ratio of the active power (P) to the apparent power (S, the product of the voltage and the current), and the value range is between 0 and 1. The closer the power factor is to 1, the more of the electric energy delivered by the power grid is converted into useful work, and the higher the electric energy utilization efficiency is.

[0108] In some embodiments, feature extraction is performed on the electrical data, including:

[0109] The total harmonic distortion rate of the electrical data is determined, and the expression of the total harmonic distortion rate is ;

[0110] The voltage deviation rate of the electrical data is determined, and the expression of the voltage deviation rate is ;

[0111] wherein THD is the total harmonic distortion rate, representing the degree of influence of the electrical data by harmonics, h is the harmonic order, H is the highest analyzed harmonic order, I h is the effective value of the hth harmonic current, I1 is the effective value of the fundamental current, is the voltage deviation rate, representing the deviation degree of the actual operating voltage from the rated voltage, is the actual voltage effective value, is the rated voltage value;

[0112] The total harmonic distortion rate and the voltage deviation rate are taken as harmonic and power quality characteristic data of the electrical data.

[0113] The total harmonic distortion rate, which is an index for describing the degree to which a waveform is affected by harmonics, and the voltage deviation rate, which is an index for quantifying the degree of deviation of an actual operating voltage of a power system from a rated value, are taken as harmonic and power quality characteristic data of electrical data.

[0114] The total harmonic distortion rate is used to quantify the difference between an actual electrical signal (voltage or current) and an ideal sinusoidal wave, usually expressed in percentage (%). It reflects the total content of each harmonic component in the signal in addition to the fundamental wave (50Hz or 60Hz). The lower the THD value, the closer the waveform is to an ideal sinusoidal wave. Harmonics are mainly generated by nonlinear loads such as frequency converters, rectifiers, electric welders, etc. Excessive THD can cause equipment to overheat, shorten its lifespan, and interfere with the normal operation of precision instruments.

[0115] The voltage deviation rate is the difference between the actual operating voltage and the system rated voltage, expressed as a percentage of the rated voltage, also expressed in percentage (%). It directly reflects the stability level of the grid voltage. For example, under a 220V rated voltage, if the actual voltage is 210V, the voltage deviation rate is about -4.5%. Excessive voltage deviation (too high or too low) can affect device performance, such as low voltage causing motor speed to drop and lighting brightness to dim, and high voltage potentially burning out electrical appliances.

[0116] In some embodiments, feature extraction is performed on the electrical data, including:

[0117] The current change rate at load startup is obtained, and the expression of the startup current change rate is ;

[0118] The current change rate at load shutdown is obtained, and the expression of the shutdown current change rate is ;

[0119] The total energy consumed or released by the load during the start-stop process is obtained, and the expression of the total energy is ;

[0120] wherein, is the startup current change rate, is the shutdown current change rate, is the current change amount, is the time window width, is the transient process energy, is the voltage data at time t, is the current data at time t, is the transient start time;

[0121] The startup current change rate, shutdown current change rate, and total energy are taken as instantaneous feature data of the electrical data.

[0122] The starting current change rate is the amount of current change per unit time during the process of current rising from the initial value (usually 0) to the stable working current. It directly reflects the current impact intensity at the moment of power-on of the circuit or device. The higher the rate, the more rapid the current rises. Excessive starting current change rate may cause instantaneous high voltage (such as the back electromotive force of inductive elements), damage semiconductor devices (such as IGBT, diode) in the circuit, or interfere with the surrounding sensitive circuits.

[0123] The shutdown current change rate is the amount of current change per unit time during the process of current falling from the stable working current to the stop value (usually 0), and the unit is consistent with the starting current change rate. It reflects the "decay speed" of current at the moment of power-off of the circuit or device. The higher the rate, the more rapid the current falls. Excessive shutdown current change rate may also cause problems, such as generating a sharp voltage in an inductive circuit, breaking down the insulation layer or damaging the switching device, and may also cause electromagnetic interference.

[0124] The total energy is the total amount of electrical energy consumed or transmitted by the circuit within a certain period of time (such as the starting phase, working phase or entire running cycle). It measures the energy scale of the electrical process, and directly corresponds to the actual energy consumption or energy transmission efficiency of the circuit. The calculation of total energy is usually related to power (P) and time (t), and can be used to evaluate the energy consumption cost of the device, the endurance capability of the battery, or the upper limit of energy storage of the energy storage element (such as capacitor, inductor).

[0125] In some embodiments, feature extraction is performed on the electrical data, including:

[0126] The skewness of the electrical data is obtained, and the expression of the skewness is ;

[0127] The index kurtosis of the electrical data is obtained, and the expression of the index kurtosis is ;

[0128] Wherein, the skewness is the skewness, which represents the asymmetry of the waveform distribution of the electrical data, the kurtosis is the index kurtosis, which represents the sensitivity to extreme values, x is the sequence of the electrical data, μ is the mean of the sequence of the electrical data, is the standard deviation of the sequence of the electrical data;

[0129] The skewness and the index kurtosis are taken as the statistical distribution features of the electrical data.

[0130] Skewness is a statistical measure of the asymmetry of the probability distribution. It can be positive, negative, or zero. Skewness = 0: The signal distribution is completely symmetric, such as the standard normal distribution, the waveform on both sides of the mean is consistent. Skewness is positive (right skew): The long tail of the signal distribution is biased to the side of the larger value, that is, most of the data is concentrated on the left side, and there are a small number of extreme large values on the right side, such as current signals containing occasional large current surges. Skewness is negative (left skew): The long tail of the signal distribution is biased to the side of the smaller value, that is, most of the data is concentrated on the right side, and there are a small number of extreme small values on the left side, such as voltage signals under certain fault conditions.

[0131] Kurtosis is a statistical measure of the steepness of the probability distribution and the thickness of the tail. It is usually compared with the kurtosis of the normal distribution (value 3). Kurtosis > 3 (spike distribution): The signal distribution is steeper than the normal distribution, the peak is more prominent, and the tail is thicker, which means that extreme values are more likely to occur in the signal, such as vibration current signals when the motor bearing fails. Kurtosis = 3 (normal distribution): The signal distribution conforms to the normal distribution, the peak and tail characteristics are moderate, such as smooth current / voltage signals during normal operation of the device. Kurtosis < 3 (flat peak distribution): The signal distribution is flatter than the normal distribution, the peak is not obvious, and the tail is thinner, which means that there are fewer extreme values in the signal, such as some filtered smooth electrical signals.

[0132] In some embodiments, before the feature fusion of the features is performed to obtain the fused data, the following steps are further included:

[0133] Normalizing the feature data of each feature;

[0134] Determining the dimension of the normalized feature data;

[0135] Extending the dimensions of all feature data to the same dimension as the feature data with the largest dimension;

[0136] Splicing the feature data after dimension expansion to obtain the fused data.

[0137] Multi-domain feature splicing: power features, harmonic and power quality indicators, instantaneous features, and statistical distribution features are spliced according to a unified format:

[0138] Each domain is spliced in a predetermined order, and the same dimension and index mapping are maintained at any time;

[0139] Z-score normalization is performed on the features within the feature domain;

[0140] Missing values are filled with 0.

[0141] Since the integrated feature dimension after splicing is high, there may be redundancy and correlation, so dimension reduction processing is needed. One way is to use principal component analysis (PCA) to calculate the feature covariance matrix and extract the first k principal components, respectively mapping the high-dimensional features of each domain to a low-dimensional space while retaining the maximum variance information.

[0142] In some embodiments, the features are fused to obtain fused data, including:

[0143] Divide the weight for each feature data;

[0144] Multiply each feature data with the corresponding weight, and accumulate the products of all feature data and the corresponding weight to obtain the fused data;

[0145] Input the fused data into the pre-trained neural network, so that the neural network outputs the load type corresponding to the electrical data, including:

[0146] Input the fused data into the pre-trained neural network, so that the neural network outputs a discriminant value;

[0147] Determine the load type based on the discriminant value.

[0148] Each type of reduced dimension feature is extracted by an independent MLP:

[0149]

[0150] wherein, and is a linear transformation weight, is an activation function, and the output dimension is unified to . refers to the normalized feature vector, t and s are indexes, the weighting coefficients of transient and steady-state features, respectively transient and steady-state feature vectors. time is the time domain, freq is the time domain, f is the time-frequency domain, and stat is the statistical feature.

[0151] Feature attention fusion:

[0152] Use a learnable attention coefficient to automatically assign the importance of each branch, calculate the attention weight i : , and fuse the features , , u i is the learnable scoring vector corresponding to the i-th feature, u j is the learnable scoring vector corresponding to the j-th feature, and iThe denominator of the feature represents the normalization of all features, j = 1, …, M.

[0153] The fused features are then mapped by a 2-3 layer fully connected network: y is the final classification category after the softmax function.

[0154] wherein, may be a ReLU function, is an output layer activation function Softmax function, y is the output category, and the specific division is performed according to the task. W1 is the weight of the first layer of the fully connected network, W2 is the weight of the second layer of the fully connected network, b1 is the bias of the first layer of the fully connected network, and b2 is the bias of the second layer of the fully connected network.

[0155] Figure 2 A structure diagram of a load discrimination device provided by the application is shown in the figure. The load discrimination device comprises:

[0156] The memory 21 is used for storing a computer program.

[0157] The processor 22 is used for executing the computer program to realize the steps of the load discrimination method.

[0158] The load discrimination device provided by the application is introduced in the above embodiment, and will not be described here.

[0159] It should be further noted that in the present specification, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the element defined by the statement "including a" does not exclude the presence of other identical elements in the process, method, article or equipment including the element.

[0160] Those skilled in the art will further realize that the mechanisms of the various examples described herein are capable of being implemented using any number of combinations of the described features. Accordingly, these examples are not limited to the mechanisms described herein, but rather, the intent is to cover all modifications and alternatives equivalent thereto. The preceding description of the examples is illustrative, and not restrictive. Many other examples will be apparent to those of skill in the art upon reviewing the above description. The scope of the examples should, therefore, be determined not with reference to the above description, but instead should be given to the appended claims, along with their full scope of equivalents.

[0161] The above description of disclosed examples is intended to be illustrative, and not restrictive. Many other examples will be apparent to those of skill in the art upon reviewing the above description. The scope of the examples should, therefore, be determined not with reference to the above description, but instead should be given to the appended claims, along with their full scope of equivalents.

Claims

1. A method for load identification, characterized in that, include: Collect electrical data from the user's main electricity meter, including voltage and current data; Feature extraction is performed on the electrical data, and the features include at least one of power features, harmonic and power quality features, instantaneous features, and statistical distribution features; The features are then fused to obtain fused data; The fused data is input into a pre-trained neural network so that the neural network outputs the load type corresponding to the electrical data, the load type including adjustable load and non-adjustable load.

2. The load identification method as described in claim 1, characterized in that, After collecting electrical data from the user's main electricity meter, the process also includes: A bandpass filter is used to filter out fundamental interference and high-frequency noise in the electrical data; The voltage data and the current data are obtained and Fourier transformed respectively to convert the voltage data and the current data from time domain data to frequency domain data; Determine the phase angle of the converted voltage and current data; Determine the phase difference between the voltage data and the current data, and the expression for the phase difference is: ; in, Let θ be the phase difference. i Let θ be the phase angle of the current data. u The phase angle of the voltage data; The phase difference is used to correct the waveform of the current data so that the waveform of the current data is aligned with the waveform of the voltage data.

3. The load identification method as described in claim 1, characterized in that, After collecting electrical data from the user's main electricity meter, the process also includes: The electrical data is segmented according to a time window of a first preset length to obtain multiple segments of electrical segmentation data; Or determine the trigger point according to preset trigger conditions; The electrical data is segmented using a time window of a second preset length where the trigger point is located, resulting in multiple segments of electrical segmentation data. Feature extraction of the electrical data includes: Feature extraction is performed on each segment of electrical data.

4. The load identification method as described in claim 1, characterized in that, Feature extraction of the electrical data includes: The active power, reactive power, and power factor of the electrical data are obtained respectively. The expression for the active power is: ; The expression for the reactive power is: ; The expression for the power factor is: ; Where P is the active power, and T is the period of a fundamental frequency. Here is the voltage data at time t. Here, S is the current data at time t, S is the apparent power, the product of the voltage data and the current data, Q is the reactive power, and PF is the power factor; The active power, the reactive power, and the power factor are used as power-type feature data of the electrical data.

5. The load identification method as described in claim 1, characterized in that, Feature extraction of the electrical data includes: Determine the total harmonic distortion rate of the electrical data, wherein the expression for the total harmonic distortion rate is: ; Determine the voltage deviation rate of the electrical data, the expression for which is: ; Wherein, THD is the total harmonic distortion rate, characterizing the degree of harmonic influence on electrical data, h is the harmonic order, H is the highest harmonic order analyzed, and I... h Ih is the effective value of the h-th harmonic charge, and I1 is the effective value of the fundamental current. The voltage deviation rate characterizes the degree of deviation between the actual operating voltage and the rated voltage. This is the actual effective value of the voltage. This is the rated voltage value; The total harmonic distortion rate and the voltage deviation rate are used as harmonic and power quality characteristic data of the electrical data.

6. The load identification method as described in claim 1, characterized in that, Feature extraction of the electrical data includes: The rate of change of the starting current is obtained, and the expression for the rate of change of the starting current is: ; Obtain the rate of change of current when the load is turned off; the expression for the rate of change of the turn-off current is: ; Obtain the total energy consumed or released by the load during start-up and shutdown, the expression for the total energy is as follows: ; in, The rate of change of the starting current. The rate of change of the turn-off current is given. The change in current The width of the time window. Energy for transient processes Here is the voltage data at time t. The current data at time t, This is the moment of transient onset; The rate of change of the starting current, the rate of change of the turning-off current, and the total energy are used as instantaneous characteristic data of the electrical data.

7. The load identification method as described in claim 1, characterized in that, Feature extraction of the electrical data includes: Obtain the skewness of the electrical data, the expression for which is: ; Obtain the index kurtosis of the electrical data, the expression for which is: ; Wherein, Skewness is the skewness, characterizing the asymmetry of the waveform distribution of the electrical data; Kurtosis is the kurtosis index, characterizing the sensitivity to extreme values; x is the sequence of the electrical data; and μ is the mean of the sequence of the electrical data. The standard deviation of the sequence of electrical data; The skewness and the index kurtosis are used as statistical distribution characteristics of the electrical data.

8. The load identification method as described in claim 1, characterized in that, Before fusing the features to obtain the fused data, the process also includes: Normalize the feature data of each feature; Determine the dimensions of the normalized feature data; Expand the dimension of all feature data to the same dimension as the feature data with the largest dimension. The feature data after dimensional expansion is concatenated to obtain fused data.

9. The method for load identification as described in any one of claims 1 to 8, characterized in that, The features are fused to obtain fused data, including: Weights are assigned to each feature data; Each feature data is multiplied by its corresponding weight, and the products of all feature data and their corresponding weights are summed to obtain the fused data. The fused data is input into a pre-trained neural network so that the neural network outputs the load type corresponding to the electrical data, including: The fused data is input into a pre-trained neural network so that the neural network outputs a discriminant value; The load type is determined based on the discriminant value.

10. A load identification device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the load identification method as described in any one of claims 1 to 9.

Images