Low-voltage transformer area power distribution network system modeling and analysis method based on KAN neural network

Abstract

The invention discloses a low-voltage transformer area power distribution network system modeling and analysis method based on a KAN neural network, and the method comprises an intelligent electric energy meter data collection module which is used for collecting the electric energy data of a measurement area and first data for model simulation building; the Simulink low-voltage transformer area power distribution network simulation model comprises a plurality of user load subsystems and a simulation model used for simulating electric energy use scenes of a plurality of users in a transformer area. Through a KAN neural network analysis method, deep learning analysis is carried out on electric energy meter data, intelligent electric energy meter errors in a transformer area and the loss of each branch line in the transformer area are calculated, electric energy data of users in the transformer area are acquired through an intelligent electric energy meter acquisition module, and load characteristics of a power utilization area are acquired through preprocessing; and building a Simulink low-voltage distribution network district system model through load characteristics and electric energy meter errors.

Description

Technical Field

[0001] The present invention relates to the technical field of modeling and analysis of low-voltage distribution network systems in an area, and in particular to a modeling and analysis method of low-voltage distribution network systems in an area based on a KAN neural network. Background Art

[0002] In low-voltage distribution substations, power transmission inevitably causes losses due to line impedance, and this loss is difficult to measure in real time. As a device used to measure and detect power loss in the power system, smart energy meters are responsible for the collection, measurement and transmission of raw power data. According to national metrology regulations, the internal components of the energy meter may cause measurement deviations due to aging, environmental changes, and technical factors during long-term operation. Therefore, the energy meter must be calibrated and rotated according to the prescribed cycle. The energy meter that does not meet the normal verification standards should be calibrated in time, and the energy meter that reaches the rotation cycle should be rotated compulsorily to ensure the accuracy, fairness and transparency of electricity metering and billing. However, according to actual conditions, most energy meters can still meet the normal verification standards when they reach the rotation cycle. Some energy meters may have abnormalities, inaccuracies and other faults during daily use. Relying only on the service life of the energy meter to determine whether the energy meter continues to be used will cause a waste of economy and resources.

[0003] Therefore, in view of this, a modeling and analysis method of low-voltage distribution network system based on KAN neural network is proposed to solve the problem that the existing substation lines are difficult to measure and the electric energy meter error is remotely calibrated. Summary of the invention

[0004] The purpose of the present invention is to provide a low-voltage distribution network system modeling and analysis method based on a KAN neural network to solve the problem of difficult measurement of distribution lines and remote calibration of electric energy meter errors proposed in the above background technology.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a low-voltage distribution network area system modeling and analysis method based on KAN neural network, comprising: a smart electric energy meter data acquisition module, a Simulink low-voltage distribution network area simulation model and a KAN neural network analysis method.

[0006] Step S1, according to the smart electric energy meter in the low-voltage distribution station area, the electric energy data of the electric energy meter measurement area is synchronously collected by the smart electric energy meter data acquisition module, wherein the first data at least includes the electric energy meter type, phase voltage, phase current, active power factor, and reactive power factor;

[0007] Step S2, preprocessing the electric energy data, and extracting the load characteristics in the metering area of ​​the smart electric energy meter through the electric energy data;

[0008] Step S3, referring to the power data obtained by each smart power meter in the area and calculating the load characteristics of the power consumption area, such as Figure 1 Build the low-voltage distribution network area system model through Simulink;

[0009] Step S4, such as Figure 2 In the low-voltage distribution network substation system model, the Simscape AC voltage source module outputs electric energy to the smart electric energy meter subsystem in the main substation area, and then flows through Figure 3 Smart electric energy meter subsystem for each branch user load subsystem;

[0010] Step S5, the low voltage distribution network area model performs simulation data, and selects continuous simulation analysis in the Powergui module;

[0011] Step S6, as Figure 4 The smart energy meter subsystem receives the continuous voltage and current vector signals of the user load subsystem, calculates the accumulated active energy of the user load subsystem through the internal power calculation module and the continuous time integrator module, and outputs the accumulated energy simulation data of the substation model calculated by each smart energy meter subsystem through the To Workspace module;

[0012] Step S7, the accumulated power simulation data is sorted according to the simulation time and screened and filtered to remove the jump power simulation data;

[0013] Step S8, the accumulated power simulation data after data processing consists of the total power of the area and the power of each user load subsystem. The relationship between the total power of the entire area and the power of each user load subsystem is:

[0014]

[0015] Where EM0(t) represents the active energy consumed by the entire substation in the tth time period (kW h), EM i (t) represents the active power consumed by the load of the ith branch in the tth time period (kW·h), n represents the number of branches included in the total meter, and L(t) represents the line loss of the entire substation in the tth time period.

[0016] During the operation of the electric energy meter, there is a measurement error. After considering the error factor, the measurement display value of the electric energy meter can be expressed as:

[0017]

[0018] in It indicates the active energy value (kW·h) displayed by the smart energy meter, EM indicates the real active energy value (kW·h) measured by the smart energy meter, and e indicates the relative error of the corresponding energy meter.

[0019] The line loss between each user load subsystem and the low voltage AC voltage source can be expressed by the user load subsystem loss electric energy:

[0020]

[0021] Where L(t) represents the total line loss in the low voltage area, α i (t) represents the proportionality coefficient between line loss and load energy loss.

[0022] The relationship between the total power of the entire area and the power of each user load subsystem can be expressed as a set of simultaneous equations:

[0023] ΓΦ0=Γ i AΦ

[0024] Where Γ represents the error coefficient of the electric energy meter in the substation, Γ i The error coefficient of the electric energy meter of each user load subsystem, Φ0 represents the electric energy value displayed by the electric energy meter in the substation, and Φ represents the electric energy value displayed by each user load subsystem.

[0025]

[0026]

[0027] in It represents the active energy value (kW·h) displayed by the energy meter, and e represents the relative error of the corresponding energy meter.

[0028] Step S9, the KAN neural network follows the Kolmogorov-Arnold theorem. If function f is a multivariate continuous function on a bounded domain, then function f can be represented by multiple single-variable functions, that is, any multivariate continuous function can be represented as a superposition of a finite number of single-variable continuous functions. That is, when there is a supervised learning task consisting of an input and output {xi, yi}, we only need to find a single-variable function Φ that satisfies the condition q and φ q,p , we can make all data points yi≈f(xi), and the relationship between input and output pairs is:

[0029]

[0030] The activation function φ(x) = w(spline(x) + b(x)), the w factor contains the sum of a basis function b(x) and a B-spline function, and the spline function splinc(x) = Σc i B i (x), c iis a trainable parameter. By parameterizing each one-dimensional function φ(x) as a B-spline curve with learnable coefficients, each data point can be interpolated or approximated.

[0031] Step S10, as Figure 5 The energy meter error and line loss analysis problem is constructed as a KAN neural network supervised learning framework, and the activation function of the KAN neural network is initialized;

[0032] Step S11, as Figure 6 The KAN neural network supervised learning model training data set is generated by low-voltage distribution network area model simulation, and the low-voltage distribution network area model simulation samples active power data every ten minutes, accumulating m groups of simulation data for n+1 days, where n represents the number of user load subsystems, and m is determined by the sampling frequency;

[0033] Step S12, the training data set is divided into a training set and a test set in a ratio of 8:2;

[0034] Step S13, the KAN neural network trains the training data set, using 500 training cycles and an Adam optimizer with a learning rate of lr=0.01 to train the activation function φ(x) of each neuron in the KAN neural network;

[0035] Step S14, the activation function φ obtained by training i (x) can be expressed by the relative error of the electric energy meter and the line loss as:

[0036]

[0037] The independent variable x represents the power loss of the user load subsystem, and the spline function Indicates the relationship between electric energy and error;

[0038] Step S15, the spline function separates the energy meter error and line loss, L = b;

[0039] Where k represents the slope of the activation function φ, and b represents the intercept of the activation function φ.

[0040] Step S16, comparing the electric energy meter error value obtained by KAN network training with the actual error value.

[0041] Preferably, in step S3, in the construction of the low-voltage distribution network area system model, the impedance characteristics of the load loop are set using the Series RLCBranch module;

[0042]

[0043] Where R represents resistance, represents the voltage vector, represents the current vector, cos(β u -β I ) represents active power factor, L represents inductance, C represents capacitance, sin(β u -β I ) represents reactive power factor;

[0044] Preferably, in step S7, data processing is performed to accumulate power simulation data, and all abnormal data at the same time need to be eliminated to ensure the time synchronization of the data and satisfy the physical relationship between the total power of the substation area and the power of each branch at the same time;

[0045] Preferably, the input and output {Xi:t, Y:t} in step S9 represent the output of each branch smart energy meter subsystem and the accumulated energy simulation data of the entire area smart energy meter subsystem;

[0046] Preferably, step S10 sets the KAN neural network supervised learning framework as a 2-layer KAN network structure, including n inputs representing n user load subsystems in the low-voltage distribution network substation area, the n neurons of the first layer of the neural network representing the error parameters and line losses of the smart power meter subsystem corresponding to each user load subsystem, and the neurons of the second layer of the neural network representing the error parameters of the smart power meter subsystem corresponding to the total substation area.

[0047] The present invention provides a method for modeling and analyzing a low-voltage distribution network system based on a KAN neural network, which has the following beneficial effects: the present invention collects the power consumption parameters of the distribution area through a smart electric energy meter, and uses Simulink to build a low-voltage distribution network distribution area system model. The distribution area model simulates the distribution area power consumption data, and the KAN neural network is used to analyze the electric energy meter errors of each user in the distribution area and the line loss of the distribution area. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is the topological structure diagram of the station area of ​​the present invention;

[0049] Figure 2 It is the Simulink low-voltage distribution network area simulation model diagram of the present invention;

[0050] Figure 3 This is a diagram of the user load subsystem of the present invention;

[0051] Figure 4 This is a subsystem diagram of the smart electric energy meter of the present invention;

[0052] Figure 5 This is a structural diagram of the KAN neural network of the present invention;

[0053] Figure 6 This is a flow chart of the KAN neural network training of the present invention. DETAILED DESCRIPTION

[0054] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following embodiments are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0055] See also Figures 1 to 6 The present invention provides a technical solution: a modeling and analysis method for a low-voltage distribution network system based on a KAN neural network, including: a smart electric energy meter data acquisition module, a Simulink low-voltage distribution network simulation model and a KAN neural network analysis method.

[0056] The smart electricity meter data acquisition module is composed of an electricity meter measurement unit, a data processing unit, a data communication unit and a concentrator. The electricity meter measurement unit is responsible for collecting current, voltage, active power and other signals in the power system on time. The data processing unit is responsible for processing the collected signals and calculating the amplitude and power factor of the voltage and current signals. The data communication unit and the concentrator are responsible for sending the processed real-time signal data to the cloud data platform.

[0057] The Simulink substation simulation model consists of a Simscape AC voltage source module and multiple user load subsystems, which are used to simulate the electricity consumption of residents in the substation area. The Simscape AC voltage source module is connected in series with the entire user load subsystem, and each user load subsystem is connected in parallel.

[0058] Where EM0(t) represents the active energy consumed by the entire substation in the tth time period (kW h), EM i (t) represents the active power consumed by the load of the ith branch in the tth time period (kW·h), n represents the number of branches included in the total meter, and L(t) represents the line loss of the entire substation in the tth time period.

[0059] The user load subsystem consists of line impedance elements, user load impedance elements, load switch elements, and smart energy meter subsystems. The line impedance elements are used to simulate the current loss generated by the lines during the power transmission process. The user load impedance elements are used to simulate the user's load usage, so that the simulation produces changes in voltage, current, and power factor. The load switch is used to control the on and off of the user load impedance element line to simulate the different usage times and frequencies of the user load. The smart energy meter subsystem is used to calculate and analyze the power consumption of the entire substation and each user load subsystem.

[0060] The smart electric energy meter subsystem consists of a voltage input port module, a current input port module, a Fourier analysis module, a power factor calculation module, a power calculation module, an error parameter module, a continuous time integrator module and a data output port module.

[0061] The high voltage end of the voltage input port module is connected to the high voltage end of the user load subsystem, the low voltage end of the voltage input port module is connected to the ground signal of the user load subsystem, and the output signal of the voltage input port module is connected to the input end of the voltage Fourier analysis module of the electric energy meter subsystem.

[0062] The current input end of the current input port module is connected to the current input end of the user load subsystem, the current output end of the current input port module is connected to the load switch element of the user load subsystem, and the signal output end of the current input port module is connected to the current Fourier analysis module input end of the electric energy meter subsystem.

[0063] The output end of the voltage Fourier analysis module is a voltage amplitude signal and a voltage phase signal, and the output end of the current Fourier analysis module is a current amplitude signal and a voltage phase signal.

[0064] The power factor calculation module makes a difference between the voltage phase signal and the current phase signal. The phase difference signal in the Simulink simulation is in angle by default. The power factor calculation module converts it into radian value, and then inputs the signal into the trigonometric function module to output the active power factor signal and the reactive power factor signal.

[0065] The input end of the power calculation module is connected to the voltage amplitude signal of the output end of the voltage Fourier analysis module, the current amplitude signal of the output end of the current Fourier analysis module, and the active power factor signal of the output end of the power factor calculation module. The active power calculation formula is:

[0066] P=UIcos(β u -β I )

[0067] Where U is the voltage across the load, I is the current in the measurement loop, β u is the voltage phase, β I is the current phase, cos(β u -β I ) calculates the load active power factor.

[0068] The error parameter module is used to simulate the error of the energy meter and multiply it with the output of the power calculation module to obtain the energy error display value. The relationship between the actual value and the display value of the energy meter error is:

[0069]

[0070] in It represents the active energy value (kW·h) displayed by the energy meter, and e represents the relative error of the corresponding energy meter.

[0071] The continuous time integrator module is used to calculate the accumulated active energy data within a certain period of time. The energy calculation formula is:

[0072]

[0073] The KAN neural network analysis method processes the electric energy data measured by the electric energy meter. The KAN neural network is suitable for multi-input single-output models, that is, any multi-variable continuous function can be expressed as a superposition of a finite number of one-variable continuous functions. The KAN neural network can represent:

[0074]

[0075] The activation function φ(x) = w(b(x) + spline(x)), the w factor contains the sum of a basis function b(x) and a B-spline function. The spline function splinc(x) = ∑c i B i (x), c i is a trainable parameter.

[0076] The input of the KAN neural network is the electric energy data output by the electric energy meter subsystem of each user load subsystem, and the output of the KAN neural network is the electric energy data output by the electric energy meter subsystem of the entire substation.

[0077] Each activation function is generated through multiple iterations of the KAN neural network model. The slope and intercept of the activation function are obtained through training. The KAN neural network training parameters are further optimized. The optimized electricity meter error and line loss results are obtained by comparing the experimental results.

[0078] The embodiments of the present invention are given for the purpose of illustration and description, and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention and to enable those of ordinary skill in the art to understand the present invention and thereby design various implementations with various modifications suitable for specific uses.

Claims

1. A method for modeling and analyzing a low-voltage distribution network system based on a KAN neural network, characterized in that: include: The smart electric energy meter data collection module is used to collect the electric energy data of the metering area and the first data for model simulation construction; The Simulink low-voltage distribution network simulation model includes multiple user load subsystems, which is used to simulate the power usage scenarios of multiple users in the distribution area; The KAN neural network analysis method is used to perform deep learning analysis on the electric energy meter data and calculate the errors of the smart electric energy meters in the substation area and the line losses of each branch line in the substation area.

2. According to a method for modeling and analyzing a low-voltage distribution network system based on a KAN neural network according to claim 1, it is characterized in that: Step S1, according to the smart electric energy meter in the low-voltage distribution station area, the electric energy data of the electric energy meter measurement area is synchronously collected by the smart electric energy meter data acquisition module, wherein the first data at least includes the electric energy meter type, phase voltage, phase current, active power factor, and reactive power factor; Step S2, preprocessing the electric energy data, and calculating the load characteristics in the metering area of ​​the smart electric energy meter through the electric energy data.

3. According to a method for modeling and analyzing a low-voltage distribution network system based on a KAN neural network according to claim 2, it is characterized in that: Step S3, referring to the electric energy data obtained by each smart electric energy meter in the substation and the calculated load characteristics, building the low-voltage substation distribution network system model through Simulink; Step S4, in the low-voltage distribution network system model, the Simscape AC voltage source module outputs electric energy to the smart electric energy meter subsystem of the main distribution area, and then flows through the smart electric energy meter subsystem of each branch user load subsystem.

4. According to a method for modeling and analyzing a low-voltage distribution network system based on a KAN neural network according to claim 3, it is characterized in that: Step S5, the low voltage distribution network model performs simulation data, and selects continuous simulation analysis in the Powergui module; Step S6, the smart energy meter subsystem receives the continuous voltage and current vector signals of the user load subsystem, calculates the accumulated active energy of the user load subsystem through the internal power calculation module and the continuous time integrator module, and outputs the accumulated energy simulation data of the substation model calculated by each smart energy meter subsystem through the ToWorkspace module; Step S7, the accumulated power simulation data is sorted according to the simulation time and screened and filtered to remove the jump power simulation data.

5. According to a method for modeling and analyzing a low-voltage distribution network system based on a KAN neural network according to claim 4, it is characterized in that: Step S8, the accumulated power simulation data after data processing consists of the total power of the area and the power of each user load subsystem. The relationship between the total power of the entire area and the power of each user load subsystem is: Where EM0(t) represents the active energy consumed by the entire substation in the tth time period (kW h), EM i (t) represents the active power consumed by the load of the i-th branch in the t-th time period (kW·h), n represents the number of branches included in the total meter, and L(t) represents the line loss of the entire substation in the t-th time period; During the operation of the electric energy meter, there is a measurement error. After considering the error factor, the measurement display value of the electric energy meter can be expressed as: in Indicates the active energy value displayed by the smart energy meter (kW·h), EM, indicates the real measured active energy value (kW·h) of the smart energy meter, and e indicates the relative error of the corresponding energy meter; The line loss between each user load subsystem and the low voltage AC voltage source can be expressed by the user load subsystem loss electric energy: Where L(t) represents the total line loss in the low voltage area, α i (t) represents the ratio coefficient between line loss and load loss electric energy; The relationship between the total power of the entire area and the power of each user load subsystem can be expressed as a set of simultaneous equations: ΓΦ0=Γ i AF Where Γ represents the error coefficient of the electric energy meter in the substation, Γ i The error coefficient of the electric energy meter of each user load subsystem, Φ0 represents the electric energy value displayed by the electric energy meter in the substation, and φ represents the electric energy value displayed by each user load subsystem; in It represents the active energy value (kW·h) displayed by the energy meter, and e represents the relative error of the corresponding energy meter.

6. According to a method for modeling and analyzing a low-voltage distribution network system based on a KAN neural network according to claim 5, it is characterized in that: Step S9, the KAN neural network follows the Kolmogorov-Arnold theorem. If the function f is a multivariate continuous function on a bounded domain, then the function f can be represented by multiple single-variable functions, that is, any multivariate continuous function can be represented as a superposition of a finite number of single-variable continuous functions, that is, when there is a supervised learning task consisting of an input and output {xi, yi}, we only need to find a single-variable function φ that satisfies the condition q and φ q，p , we can make all data points yi≈f(xi), and the relationship between input and output pairs is: The activation function φ(x) = w(spline(x) + b(x)), the w factor contains the sum of a basis function b(x) and a B-spline function, and the spline function splinc(x) = Σc i B i (x), c i It is a trainable parameter. By parameterizing each one-dimensional function φ(x) into a B-spline curve with learnable coefficients, each data point can be interpolated or approximated.

7. According to claim 6, a method for modeling and analyzing a low-voltage distribution network system based on a KAN neural network is characterized in that: Step S10, the energy meter error and line loss analysis problem is constructed as a KAN neural network supervised learning framework, and the activation function of the KAN neural network is initialized; Step S11, the KAN neural network supervised learning model training data set is generated by low-voltage distribution network model simulation. The low-voltage distribution network model simulation samples active power data every ten minutes and accumulates n+1 days of simulation data, where n represents the number of user load subsystems.

8. According to the method for modeling and analyzing a low-voltage distribution network system based on a KAN neural network according to claim 7, it is characterized in that: Step S12, the training data set is divided into a training set and a test set in a ratio of 8:2; Step S13, the KAN neural network trains the training data set, using 500 training cycles and an Adam optimizer with a learning rate of lr=0.01 to train the activation function φ(x) of each neuron in the KAN neural network; Step S14, the activation function φ obtained by training i (x) can be represented by the relative error of the electric energy meter and the line loss.

9. The method for modeling and analyzing a low-voltage distribution network system based on a KAN neural network according to claim 8 is characterized in that: Step S15, the spline function separates the electric energy meter error and line loss; Step S16, comparing the electric energy meter error value obtained by KAN network training with the actual error value.

Images