A method, system, equipment, medium, and product for identifying the operating status of a distributed photovoltaic (PV) distribution area.

Abstract

This invention discloses a method, system, device, medium, and product for identifying the operating status of distributed photovoltaic (PV) power distribution areas. It relates to the field of power distribution area status monitoring technology. The method involves collecting three-phase electrical parameters at the common coupling point of the distributed PV power distribution area and performing Clarke transform. A complex time-frequency matrix is ​​obtained using a polynomial adaptive S-transform, and a time-time transform matrix is ​​obtained using a discrete inverse Fourier transform. Based on these matrices, time-domain, frequency-domain, and time-time multi-dimensional features are extracted. An initial power distribution area operating status identification model is then constructed. The model's hyperparameters are optimized using a moth phototaxis algorithm that incorporates chaotic mapping to initialize the population and iterative energy switching strategies to obtain a target power distribution area operating status identification model. Finally, the time-frequency multi-dimensional features are input into the target model to achieve status identification, improving the accuracy of operating status identification for distributed PV power distribution areas under complex conditions such as reverse power transmission and islanded operation.

Description

Technical Field

[0001] This invention relates to the field of transformer substation status monitoring technology, and in particular to a method, system, equipment, medium, and product for identifying the operating status of distributed photovoltaic transformer substations. Background Technology

[0002] With the rapid development of distributed photovoltaic (PV) power generation technology, its application in distribution networks is becoming increasingly widespread. The integration of distributed PV power generation has altered the operational characteristics of traditional distribution networks, posing new challenges to the intelligent identification of transformer substation operating status. Especially in scenarios such as equipment maintenance and fault handling, the identification of various operating statuses under complex conditions such as reverse power transmission and islanded operation directly impacts the safety of workers. Therefore, there is an urgent need for an intelligent identification method tailored to the needs of operational safety monitoring, providing reliable safety protection for on-site operations while ensuring the safe and stable operation of the power grid.

[0003] Traditional identification methods have significant shortcomings in the field of operational safety monitoring: passive detection technology identifies the status by comparing electrical parameters at a common coupling point with preset thresholds, but there are blind spots when system parameters change slightly, and the subjective dependence of threshold setting leads to deviations in sensitivity and accuracy, and it cannot dynamically adapt to the differentiated characteristics of different distribution areas; active detection methods trigger inverter responses by injecting disturbance signals, but injected interference can easily cause system oscillations and reduce power quality; remote solutions rely on the continuity of communication signals for discrimination, require the deployment of a large number of terminal devices, and are susceptible to interference from complex environments, resulting in status identification failure when communication is interrupted. The above technologies have equipment modification costs and cannot dynamically capture status change characteristics under complex operating conditions (such as reverse power supply, islanded operation, etc.), and cannot simultaneously achieve non-intrusive detection, threshold-free operation, and strong anti-interference capabilities. They are difficult to meet the non-intrusive, threshold-free distributed photovoltaic distribution area operational status identification requirements for operational safety monitoring, ultimately resulting in insufficient identification efficiency for dangerous operating conditions such as reverse power supply and islanded operation, causing maintenance personnel to face the risk of electric shock or misoperation. Summary of the Invention

[0004] This invention provides a method, system, device, medium, and product for identifying the operating status of distributed photovoltaic power stations, solving the technical problem of how to improve the accuracy of identifying the operating status of distributed photovoltaic power stations under complex operating conditions.

[0005] The first aspect of this invention provides a method for identifying the operating status of a distributed photovoltaic power station, comprising:

[0006] The three-phase electrical parameters at the common coupling point of the distributed photovoltaic system are collected and subjected to Clarke transformation to obtain the Clarke transformation components.

[0007] The time-frequency transformation matrix is ​​determined based on the complex time-frequency matrix obtained by adaptive S-transform of the Clarke transform components.

[0008] Multivariate feature extraction is performed using the complex time-frequency matrix and the time-time transformation matrix to obtain time-frequency multivariate features;

[0009] An initial transformer area operation status identification model is constructed, and the parameters of the initial transformer area operation status identification model are optimized by the noctuid moth phototaxis algorithm to obtain the target transformer area operation status identification model.

[0010] The time-frequency multivariate features are input into the target transformer area operation status identification model for status identification, and the transformer area operation status identification result is obtained.

[0011] Optionally, the step of determining the time-frequency transformation matrix based on the complex time-frequency matrix obtained by adaptive S-transform of the Clarke transform components includes:

[0012] Based on the Gaussian window function, a polynomial adaptive S-transform is performed on the Clarke transform components to obtain a complex time-frequency matrix;

[0013] Perform a discrete inverse Fourier transform on the complex time-frequency matrix to obtain the time-time transform matrix.

[0014] Optionally, the time-frequency multivariate features include time-domain features, frequency-domain features, and time-time-domain features. The step of extracting multivariate features using the complex time-frequency matrix and the time-time transform matrix to obtain the time-frequency multivariate features includes:

[0015] Taking the modulus of the complex time-frequency matrix yields the modulus time-frequency matrix;

[0016] The time-domain features are obtained by using the modulus-time-frequency matrix.

[0017] Frequency domain features are obtained by using the modulus-time-frequency matrix;

[0018] The time-domain features are obtained by using the time-time transformation matrix.

[0019] Optionally, the step of constructing an initial transformer area operation status identification model and optimizing the parameters of the initial transformer area operation status identification model using a moth phototaxis algorithm to obtain a target transformer area operation status identification model includes:

[0020] Construct an initial transformer area operation status identification model and determine the hyperparameter search range;

[0021] A chaotic sequence of random numbers is generated using chaotic mapping, and the chaotic sequence of random numbers is mapped to the hyperparameter search range to obtain the initial noctuid moth population;

[0022] Each noctuid moth in the initial noctuid moth population corresponds to a set of model hyperparameter combinations, and each set of model hyperparameter combinations represents the initial position of the noctuid moth for the corresponding noctuid moth individual.

[0023] Based on a pre-constructed sample set, the initial operating status identification model of the transformer area is trained using the model hyperparameter combination corresponding to each individual moth, and the model fitness of each individual moth is determined.

[0024] Based on the model fitness, each moth individual is sorted in descending order, and a preset number of moth individuals are selected from largest to smallest as the initial light source positions.

[0025] The flight speed of each individual moth is initialized using the hyperparameter search range as a constraint.

[0026] The iteration energy is calculated based on the current iteration number and the preset maximum iteration number, and then compared with the iteration energy and the preset energy threshold.

[0027] When the iteration energy is greater than or equal to the preset energy threshold, the preset global search strategy is used as the target position update strategy; when the iteration energy is less than the preset energy threshold, the preset local optimization strategy is used as the target position update strategy.

[0028] Based on the target position update strategy, and combining the initial light source position, the initial position of the moth, and the initialized flight speed, the positions of all individual moths are updated to obtain a new moth population.

[0029] Based on the new noctuid moth population, the process jumps to the step based on the pre-divided training sample set until the current iteration number equals the preset maximum iteration number. Then, the set of model hyperparameters with the highest current model fitness is selected as the optimal model hyperparameters combination.

[0030] Substituting the optimal model hyperparameter combination into the initial transformer area operation status identification model yields the target transformer area operation status identification model.

[0031] Optionally, the preset global search strategy specifically involves shifting and updating the model hyperparameters corresponding to individual moths.

[0032] Optionally, the preset local optimization strategy specifically involves fine-tuning and updating the model hyperparameters corresponding to individual moths.

[0033] A second aspect of the present invention provides a distributed photovoltaic power station operation status identification system, comprising:

[0034] The conversion module is used to collect the three-phase electrical parameters at the common coupling point of the distributed photovoltaic system and perform Clarke transformation to obtain Clarke transformation components.

[0035] The analysis module is used to determine the time-to-time transformation matrix based on the complex time-frequency matrix obtained by adaptive S-transform of the Clarke transform components;

[0036] The extraction module is used to perform multivariate feature extraction using the complex time-frequency matrix and the time-time transformation matrix to obtain time-frequency multivariate features;

[0037] The optimization module is used to construct an initial transformer area operation status identification model and optimize the parameters of the initial transformer area operation status identification model using the noctuid moth phototaxis algorithm to obtain the target transformer area operation status identification model.

[0038] The identification module is used to input the time-frequency multi-dimensional features into the target area operation status identification model for status identification, and obtain the operation status identification result of the area.

[0039] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the distributed photovoltaic area operation status identification method as described above.

[0040] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the distributed photovoltaic area operation status identification method as described above.

[0041] The fifth aspect of the present invention provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer performs the distributed photovoltaic area operation status identification method as described above.

[0042] As can be seen from the above technical solutions, the present invention has the following advantages:

[0043] This invention provides a distributed photovoltaic (PV) transformer area operation status identification scheme that requires neither injecting disturbance signals into the system nor manually setting identification thresholds. It collects three-phase electrical parameters at the common coupling point of the distributed PV system, performs Clarke transform, and then uses a polynomial adaptive S-transform to obtain a complex time-frequency matrix. This is followed by a discrete inverse Fourier transform to obtain a time-time transformation matrix. Based on these matrices, time-domain, frequency-domain, and time-time multi-dimensional features are extracted. An initial transformer area operation status identification model is then constructed. The model's hyperparameters are optimized using a moth phototaxis algorithm that incorporates chaotic mapping to initialize the population and iterative energy switching strategies to obtain a target transformer area operation status identification model. Finally, the time-frequency multi-dimensional features are input into the target model to achieve status identification. The adaptive S-transform can adapt to complex operating conditions such as reverse power transmission and islanded operation of distributed PV transformer areas. Obtaining the complex time-frequency matrix based on Clarke transform components and determining the time-time transformation matrix can adaptively match the time-frequency analysis resolution, accurately capturing instantaneous changes and frequency characteristics of the signal, thus providing a basis for multi-dimensional feature extraction. It provides a high-quality analytical foundation; multivariate feature extraction based on complex time-frequency matrices and time-to-time transformation matrices can comprehensively acquire time-frequency multivariate features, characterizing signal features under various operating conditions without setting thresholds, thus avoiding the subjectivity and detection blind spots of traditional threshold identification; when optimizing the parameters of the initial transformer area operation status identification model using the moth phototaxis algorithm, the initial position of the moth generated by chaotic mapping ensures the ergonomics of the population, and the selected initial light source position provides a benchmark for optimization. Combined with the flight speed adjusted by the iterative energy switching strategy, it can adaptively switch between global search strategy and local optimization strategy, achieving a dynamic balance between global exploration and local development during hyperparameter optimization. The synergistic effect of the three significantly improves the optimization effect of the model hyperparameters, enabling the optimized target model to fully learn the multivariate feature patterns under complex operating conditions. Ultimately, it greatly improves the accuracy of distributed photovoltaic transformer area operation status identification under complex operating conditions such as reverse power transmission and islanded operation, effectively solving the problem of insufficient identification efficiency under complex operating conditions in existing technologies. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating the steps of a method for identifying the operating status of a distributed photovoltaic power station area according to Embodiment 1 of the present invention.

[0046] Figure 2 This is a flowchart illustrating the steps of a method for identifying the operating status of a distributed photovoltaic power station area according to Embodiment 2 of the present invention.

[0047] Figure 3 This is a schematic diagram of the iterative error convergence curve of the phototaxis algorithm for moths provided in Embodiment 2 of the present invention.

[0048] Figure 4 This is a schematic diagram of the test set confusion matrix of the target transformer area operation status identification model provided in Embodiment 2 of the present invention;

[0049] Figure 5 This is a structural block diagram of a distributed photovoltaic power station operation status identification system provided in Embodiment 3 of the present invention;

[0050] Figure 6 This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention. Detailed Implementation

[0051] This invention provides a method, system, device, medium, and product for identifying the operating status of distributed photovoltaic (PV) power grids, which addresses the technical problem of improving the accuracy of identifying the operating status of distributed PV power grids under complex operating conditions.

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] The purpose of this invention is to provide an intelligent identification scheme for the operating status of distributed photovoltaic power stations for operational safety monitoring, which can accurately identify the different operating statuses of the power station under complex operating conditions (reverse power supply condition, no reverse power supply condition, islanded operation, and normal operation) and overcome the above-mentioned technical problems.

[0054] Please see Figure 1 , Figure 1 The flowchart illustrates the steps of a method for identifying the operating status of a distributed photovoltaic power station area, as provided in Embodiment 1 of the present invention.

[0055] This invention provides a method for identifying the operating status of a distributed photovoltaic (PV) distribution area, comprising:

[0056] Step 101: Collect the three-phase electrical parameters at the common coupling point of the distributed photovoltaic system and perform Clark transformation to obtain the Clark transformation components.

[0057] In this embodiment of the invention, the three-phase voltage and three-phase current electrical parameters at the common coupling point of the distributed photovoltaic system are collected, the three-phase electrical parameters are discretely sampled, and the sampled three-phase electrical parameters are subjected to Clarke transform operation to obtain the corresponding Clarke transform components.

[0058] Step 102: Determine the time-frequency transformation matrix based on the complex time-frequency matrix obtained by adaptive S-transform of the Clark transform components.

[0059] In this embodiment of the invention, the Clark transform components are analyzed in the time-frequency domain using a polynomial adaptive S-transform to calculate a complex time-frequency matrix that characterizes the time-frequency distribution characteristics of the signal; a discrete inverse Fourier transform operation is performed on the complex time-frequency matrix to obtain a time-time transform matrix that characterizes the time-domain correlation characteristics of the signal.

[0060] Step 103: Use the complex time-frequency matrix and the time-time transformation matrix to perform multivariate feature extraction to obtain time-frequency multivariate features.

[0061] In this embodiment of the invention, a modulo operation is performed on the complex time-frequency matrix to obtain a modulo time-frequency matrix. Based on the modulo time-frequency matrix, the time-domain features and frequency-domain features of the signal are extracted respectively. Based on the time-time transformation matrix, the time-time-domain features of the signal are extracted. The extracted time-domain, frequency-domain, and time-time-domain features are fused to obtain time-frequency multi-dimensional features.

[0062] Step 104: Construct an initial transformer area operation status identification model, and optimize the parameters of the initial transformer area operation status identification model using the moth phototaxis algorithm to obtain the target transformer area operation status identification model.

[0063] In this embodiment of the invention, an initial transformer substation operation status identification model based on a Light Gradient Boosting Machine (LightGBM) is constructed to determine the range of values ​​for the hyperparameters to be optimized. A phototaxis algorithm for moths is used to optimize the hyperparameters. An initial moth population is generated through chaotic mapping to determine the initial positions of the moths. The model fitness of each individual moth is calculated, and the initial light source position is selected. Simultaneously, the moth flight speed is initialized. The moth positions are updated based on iterative energy dynamic switching between a global search strategy and a local optimization strategy. The optimal hyperparameter combination is obtained after iterative optimization until the termination condition is met. The optimal hyperparameters are then substituted into the initial transformer substation operation status identification model to obtain the optimized target transformer substation operation status identification model.

[0064] Step 105: Input the time-frequency multi-dimensional features into the target area operation status identification model to perform status identification and obtain the operation status identification result of the area.

[0065] In this embodiment of the invention, the time-frequency multi-dimensional features obtained by processing the three-phase electrical parameters of the distributed photovoltaic common coupling point to be identified are input into the target area operation status identification model. The model classifies and judges the time-frequency multi-dimensional features and finally outputs the operation status identification result corresponding to the distributed photovoltaic area.

[0066] Distributed photovoltaic common coupling point: The common connection node between the distributed photovoltaic grid-connected line and the distribution area power grid, serving as the acquisition location for three-phase electrical parameters; Three-phase electrical parameters: The three-phase voltage and three-phase current signals acquired at the distributed photovoltaic common coupling point; Clarke transform: A linear transformation method that converts three-phase electrical parameters in a three-phase stationary coordinate system into components in a two-phase stationary coordinate system; Clarke transform components: The α, β, and zero-sequence components obtained after the Clarke transform of the three-phase electrical parameters; Adaptive S-transform: An improved S-transform time-frequency analysis method that achieves adaptive resolution adjustment based on a Gaussian window function; Complex time-frequency matrix: A complex time-frequency matrix containing amplitude and phase information obtained after the Clarke transform components undergo an adaptive S-transform; Time-time transformation matrix: Obtained by performing a discrete inverse Fourier transform on the complex time-frequency matrix, used for... The system comprises: a matrix representing the correlation characteristics across different time dimensions; time-frequency multivariate features: a comprehensive feature vector integrating time-domain features, frequency-domain features, and time-time-domain features; initial transformer substation operation status identification model: a basic identification model for classifying transformer substation operation status without hyperparameter optimization; moth phototaxis algorithm: an intelligent optimization algorithm for iteratively optimizing the hyperparameters of the initial transformer substation operation status identification model; parameter optimization: the process of searching for the optimal parameter combination within the hyperparameter range using the moth phototaxis algorithm; target transformer substation operation status identification model: the final usable identification model obtained after parameter optimization of the initial transformer substation operation status identification model using the moth phototaxis algorithm; transformer substation operation status identification result: the output result of the target transformer substation operation status identification model after classifying the input time-frequency multivariate features.

[0067] This invention provides a distributed photovoltaic (PV) transformer area operation status identification scheme that requires neither injecting disturbance signals into the system nor manually setting identification thresholds. It collects three-phase electrical parameters at the common coupling point of the distributed PV system, performs Clarke transform, and then uses a polynomial adaptive S-transform to obtain a complex time-frequency matrix. This matrix is ​​then further processed by a discrete inverse Fourier transform to obtain a time-to-time transform matrix. Based on these matrices, time-domain, frequency-domain, and time-domain multi-dimensional features are extracted. An initial transformer area operation status identification model is then constructed. The model's hyperparameters are optimized using a moth phototaxis algorithm that incorporates chaotic mapping to initialize the population and iterative energy switching strategies to obtain a target transformer area operation status identification model. Finally, the time-frequency multi-dimensional features are input into the target model to achieve status identification. The polynomial adaptive S-transform optimizes window function parameters for complex operating conditions such as reverse power supply and islanded operation, achieving adaptive time-frequency resolution. This provides high time resolution for disturbance signals to capture instantaneous changes and high frequency resolution for steady-state signals to detect frequency features. The extraction of highly discriminative time-frequency features, along with multi-dimensional feature extraction in the time, frequency, and time domains, effectively captures the patterns, energy, and trends of the operating signals of the distribution area. It comprehensively characterizes signal differences under different operating conditions without relying on threshold settings, avoiding the subjectivity and blind spots of traditional threshold methods. In the moth phototaxis algorithm, the initial moth position generated based on chaotic mapping ensures the traversal of the initial population, laying the foundation for global search. The initial light source position, serving as a high-quality solution reference, provides stable directional guidance for hyperparameter optimization. Combined with the dynamically adjusted flight speed using an iterative energy switching strategy, it achieves an effective balance between global exploration and local development. The synergistic effect of these three elements significantly improves the optimization effect of the model's hyperparameters, enabling the optimized target model to fully learn multi-feature patterns under complex operating conditions. Ultimately, this greatly improves the accuracy of identifying the operating status of distributed photovoltaic distribution areas under complex conditions such as reverse power transmission and islanded operation, effectively solving the problem of insufficient identification efficiency under complex conditions in existing technologies.

[0068] Please see Figure 2 , Figure 2 The flowchart illustrates the steps of a method for identifying the operating status of a distributed photovoltaic power station area, as provided in Embodiment 2 of the present invention.

[0069] This invention provides a method for identifying the operating status of a distributed photovoltaic (PV) distribution area, comprising:

[0070] Step 201: Collect the three-phase electrical parameters at the common coupling point of the distributed photovoltaic system and perform Clark transformation to obtain the Clark transformation components.

[0071] In this embodiment of the invention, in a distribution area considering grid-connected operation of distributed photovoltaic (PV) power plants, three-phase electrical parameters at the point of common coupling (PCC) of the distributed PV power plants are collected. These three-phase electrical parameters include the three-phase voltage U. a U b Uc and three-phase current I a I b I c Let N be the number of sampling points for each phase voltage and current, and the sampling rate be f. s =1 / t, sampling interval is t; Clarke transform is performed on the collected three-phase voltage and three-phase current respectively to obtain the corresponding Clarke transform components U. α U β 、U0 and I α I β 、I0.

[0072] The Clarke transform formula for the three-phase voltage is as follows:

[0073]

[0074] In the formula, These are the three-phase voltages obtained after Clark transformation. Axial components, Axis components and zero axis components; The acquired three-phase voltage signal is a component of the three-phase electrical parameters.

[0075] The Clarke transform formula for three-phase current is:

[0076]

[0077] In the formula, These are the three-phase currents obtained after Clarke transformation. Axial components, Axis components and zero axis components; The acquired three-phase current signal is a component of the three-phase electrical parameters.

[0078] Step 202: Based on the Gaussian window function, perform a polynomial adaptive S-transform on the Clark transform components to obtain the complex time-frequency matrix.

[0079] In this embodiment of the invention, the Clarke transform component includes voltage. Quantity Voltage zero-sequence component Current Quantity and the zero-sequence component of the current ,in, The active power component can reflect the dynamic changes in the active power of the system and is highly sensitive to voltage and power fluctuations during the operation of the transformer substation. The zero-sequence component can reflect the unbalanced state of the three-phase system and is an important indicator for evaluating the stability of the system operation.

[0080] Using a preferred Gaussian window function, a polynomial adaptive S-transform is performed on the Clark transform components to achieve adaptive time-frequency resolution analysis and obtain the corresponding complex time-frequency matrix.

[0081] The preferred expression for the Gaussian window function is:

[0082]

[0083] In the formula, The preferred standard deviation control function; The polynomial control adjustment parameters for the window function; Sampling time; The Gaussian window is designed for frequency detection. It has variable resolution characteristics, with the height and width of the window changing with the detection frequency. As the detection frequency increases, the height of the window gradually decreases and the width gradually increases, thus gradually improving the frequency resolution.

[0084] The continuous form of the polynomial adaptive S-transform is expressed as:

[0085]

[0086] In the formula, The frequency shift result of the Fourier transform spectrum of the Clark transform component signal; It is a time variable; For frequency variables; The imaginary unit; These are polynomial control adjustment parameters consistent with the Gaussian window function.

[0087] By variable substitution, let The formula for calculating the polynomial adaptive discrete S-transform is obtained as follows:

[0088]

[0089] In the formula, The frequency shift result of the discrete Fourier transform spectrum of the Clark transform component signal; For discrete variables, the values ​​are: ; This represents the number of sampling points; The sampling interval; The imaginary unit; The polynomial control adjustment parameters for the window function are set to a=0.0007, b=0.5, c=0.0003, and d=0.04 respectively. This is the complex time-frequency matrix obtained after the polynomial adaptive discrete S-transform, where the column vectors of the matrix are the expansion results of all frequencies corresponding to a certain time, and the row vectors of the matrix are the expansion results of all time points corresponding to a certain frequency.

[0090] Among them, frequency shift results The formula for calculation is:

[0091]

[0092] In the formula, The discrete form of the Clarke transform component signal (i.e., voltage) or current Discrete sampling sequence); For discrete variables, the values ​​are: ; This represents the number of sampling points; It is the imaginary unit.

[0093] Step 203: Perform a discrete inverse Fourier transform on the complex time-frequency matrix to obtain the time-to-time transform matrix.

[0094] In this embodiment of the invention, to further analyze the Clarke transform components (voltage) Components, zero-sequence voltage component, current The characteristics of each frequency component (including the zero-sequence current component) are analyzed to improve the time resolution of signal analysis and characterize the time local characteristics of the signal. The complex time-frequency matrix obtained by the polynomial adaptive discrete S-transform in step 202 is subjected to discrete inverse Fourier transform to calculate the time-time transform matrix. This matrix is ​​used to characterize the amplitude distribution of various frequency components of the signal in the time-time domain.

[0095] The formula for calculating the time-to-time transform matrix using the discrete inverse Fourier transform is:

[0096]

[0097] In the formula, The time-to-time transformation matrix represents the signal at discrete time points. and The connection; The complex time-frequency matrix obtained by the polynomial adaptive discrete S-transform in step 202; For discrete variables, the values ​​are: ; This represents the number of sampling points; The sampling interval; It is the imaginary unit.

[0098] Gaussian window function: A window function used in polynomial adaptive S-transform that can adaptively adjust the window width and height with frequency; Polynomial adaptive S-transform: An improved S-transform based on Gaussian window function, applied to Clarke transform components, and possessing adaptive adjustment capability for time-frequency resolution; Discrete inverse Fourier transform: A discrete signal transformation algorithm that maps and transforms the complex time-frequency matrix from the time-frequency domain to the time-time domain, and then solves for the time-time transform matrix.

[0099] Step 204: Use the complex time-frequency matrix and the time-time transformation matrix to perform multivariate feature extraction to obtain time-frequency multivariate features.

[0100] Furthermore, the time-frequency multivariate features include time-domain features, frequency-domain features, and time-time-domain features. Step 204 may include the following sub-steps:

[0101] S11. Take the modulus of the complex time-frequency matrix to obtain the modulo time-frequency matrix.

[0102] In this embodiment of the invention, in order to extract the signal features of different operating states of the transformer area, the complex time-frequency matrix obtained by the polynomial adaptive discrete S-transform in step 202 is subjected to modulo operation to obtain the corresponding modulo time-frequency matrix.

[0103] The formula for modulo operation is:

[0104]

[0105] In the formula, The modulus-time-frequency matrix obtained by the operation; For discrete variables, the values ​​are: ; This represents the number of sampling points; The sampling interval; It is the imaginary unit.

[0106] The modulo-time-frequency matrix can reflect the energy distribution and variation of the signal in the time and frequency domain, providing reliable data support for extracting the time and frequency domain characteristics of the operating status of the transformer substation.

[0107] It should be noted that the modulo-time-frequency matrix can reflect the energy distribution and variation law of the signal in the time-frequency domain, while the time-time transformation matrix describes the dynamic variation characteristics of the signal in the time-time domain. By analyzing the modulo-time-frequency matrix and the time-time transformation matrix, we can extract the multi-dimensional characteristics of the station's operating status in the time domain, frequency domain, and time-time domain, thereby effectively capturing important information in the signal, including signal mode, energy characteristics, and time series variation trends, providing a reliable basis for accurate identification of the station's operating status.

[0108] S12. Use the modulus-time-frequency matrix to extract time-domain features and obtain time-domain features.

[0109] In this embodiment of the invention, based on the modulus-time-frequency matrix, time-domain features of multiple dimensions are calculated and extracted sequentially. The specific process is as follows:

[0110] First, calculate the maximum and minimum values ​​of each row in the modulus-time-frequency matrix to obtain the maximum value vector. and minimum value vector ; where the maximum value vector The Vector of elements and minimum value The The calculation formulas for each element are as follows:

[0111]

[0112]

[0113] In the formula, Maximum value vector The One element; Minimum value vector The Each element.

[0114] Secondly, calculate the maximum vector. with minimum value vector The correlation coefficient, as the first time-domain feature, is calculated as follows:

[0115]

[0116] In the formula, This is the first time-domain feature; Maximum value vector The average value; Minimum value vector The average value; this correlation coefficient can reflect the correlation of the signal change trend in the photovoltaic area and highlight the time domain characteristics of the signal.

[0117] Next, based on the fundamental frequency amplitude vector of the modulo-time-frequency matrix, the fundamental frequency time-domain characteristics of the transformer area's operating state are characterized. The formula for calculating the fundamental frequency amplitude vector is:

[0118]

[0119] In the formula, This is the fundamental frequency amplitude vector; These are the elements corresponding to the fundamental frequency component in the modulo-time-frequency matrix; It is the fundamental frequency component; This is the discrete frequency index corresponding to the fundamental frequency.

[0120] Then, the average value of the fundamental frequency time amplitude curve is calculated as the second time-domain feature, and the calculation formula is:

[0121]

[0122] In the formula, This is the second time-domain feature, reflecting the central trend of signal disturbance changes; The first of the fundamental frequency amplitude vectors Each element.

[0123] Finally, the standard deviation of the amplitude of the fundamental frequency time-amplitude curve is calculated as the third time-domain characteristic. The calculation formula is as follows:

[0124]

[0125] In the formula, This is the third time-domain feature, used to evaluate the dispersion of the signal's time-domain feature data.

[0126] The above obtained , , Together, they constitute the temporal domain characteristics, which can effectively capture the temporal patterns and trends of signals.

[0127] S13. Frequency domain features are extracted using the modulus-time-frequency matrix to obtain frequency domain features.

[0128] In this embodiment of the invention, the maximum value of each column of the modulus-time-frequency matrix is ​​taken according to the frequency dimension to calculate the maximum frequency amplitude vector. Each element of this vector corresponds to the maximum amplitude at a discrete frequency point. Arranging these elements in frequency order creates the discrete data points of the frequency amplitude envelope curve, calculated as follows:

[0129]

[0130] In the formula, It is the vector of maximum frequency amplitude (i.e., the discrete data of the frequency amplitude envelope curve). It is a discrete-time variable, and its value range is... ; For discrete frequency variables, the value is... ; This represents the number of sampling points.

[0131] Secondly, based on the frequency amplitude envelope curve, the three largest peak values ​​and their corresponding frequency values ​​are extracted as frequency domain features. These features can reflect the main frequency components and primary and secondary characteristics of signal disturbances under different operating conditions of the transformer area. Specifically, the peaks of the frequency amplitude envelope curve are first identified using a peak detection function to obtain each peak value and its corresponding frequency value; then, the peak values ​​are sorted from largest to smallest, and the top three peak values ​​are selected. , , and their corresponding frequency values , , The corresponding formula for calculating the frequency domain features is:

[0132]

[0133] In the formula, , , These are the frequency domain characteristics corresponding to the first three peak values; , , These are the frequency values ​​corresponding to the first three peaks; For the first Each wave peak; For the first The frequency values ​​corresponding to each peak. ; For peak detection function; This is a sorting function based on peak value from largest to smallest.

[0134] The above extraction results , , , , , Together, they constitute the frequency domain characteristics, which can effectively characterize the frequency domain disturbance characteristics of the operating status of the transformer area.

[0135] S14. Use the time-time transformation matrix to extract time-time domain features and obtain time-time domain features.

[0136] In this embodiment of the invention, based on the time-to-time transformation matrix obtained in step 203, the time-to-time domain features of the station's operating status are extracted. The specific process is as follows:

[0137] First, extract the diagonal elements of the time-to-time transformation matrix. Since the diagonal elements of the time-to-time transformation matrix obtained by the discrete inverse Fourier transform after the polynomial adaptive S-transform contain key information about high-frequency components, we first extract the diagonal elements of the time-to-time transformation matrix. The calculation formula is:

[0138]

[0139] In the formula, This is the diagonal element vector of the time-varying transformation matrix; This is a function for extracting diagonal elements of a matrix. The time-to-time transformation matrix obtained in step 203; For discrete indices of diagonal elements, the value is... .

[0140] Secondly, based on the diagonal element vector, its mean and standard deviation are calculated as time-domain features of the high-frequency components: the mean of the diagonal elements (denoted as the 10th time-domain feature) is calculated as follows:

[0141]

[0142] In the formula, The mean characteristic of the diagonal elements; The first element of the diagonal vector One element; This represents the number of sampling points; For discrete indexes, the value is... .

[0143] The standard deviation of the diagonal elements (denoted as the 11th time-domain feature) is calculated as follows:

[0144]

[0145] In the formula, The standard deviation of the diagonal elements is used to assess the dispersion of the diagonal elements.

[0146] Next, contour features of the time-to-time transformation matrix are extracted. The contour lines of the time-to-time transformation matrix reflect the frequency localization of the time series in the time-time domain. To extract contour features of the operating status of the transformer substations, contour values ​​are uniformly sampled and selected within the global scope of the time-to-time transformation matrix to construct a contour set.

[0147]

[0148] Among them, the Contour values The formula for calculation is:

[0149]

[0150] In the formula, It is a set of contour line values; For the first Contour values; This is the preset number of contour line samples; This represents the minimum value of the time-to-time transformation matrix; This represents the maximum value of the time-varying transformation matrix; For contour indexing, .

[0151] Then, the contour energy density corresponding to each contour value is calculated, which serves as the contour feature of the time-to-time transformation matrix. The energy density corresponding to the first contour line value (denoted as the i-th contour line value) The formula for calculating the time-domain features is:

[0152]

[0153] In the formula, For the first Energy density characteristics corresponding to each contour line value; This is an indicator function.

[0154] Among them, indicator function The definition of is:

[0155]

[0156] In the formula, when the input variable When the sum is 0, the function value is 1; otherwise, the function value is 0. This is used to count the values ​​in the time-to-time transformation matrix. The number of elements.

[0157] The above extraction results , and the energy density characteristics of each contour line ( These factors together constitute the time-domain characteristics, which can effectively characterize the dynamic changes in the operating status of the transformer area in the time domain.

[0158] Time-domain features: Features extracted from the modulo-time-frequency matrix that characterize the time-dimensional variation of a signal; Frequency-domain features: Features extracted from the modulo-time-frequency matrix that characterize the frequency-dimensional distribution of a signal; Time-time-domain features: Features extracted from the time-time transformation matrix that characterize the time-related properties of a signal; Modulo-time-frequency matrix: A real matrix that retains only the signal amplitude energy distribution information after obtaining the modulus of the complex time-frequency matrix.

[0159] Step 205: Construct an initial transformer area operation status identification model, and optimize the parameters of the initial transformer area operation status identification model using the moth phototaxis algorithm to obtain the target transformer area operation status identification model.

[0160] Furthermore, step 205 may include the following sub-steps:

[0161] S21. Construct an initial transformer area operation status identification model and determine the hyperparameter search range.

[0162] In this embodiment of the invention, in order to balance accuracy and operational efficiency, an initial transformer area operation status recognition model based on a Light Gradient Boosting Machine (LightGBM) is constructed for multi-class recognition of transformer area operation status. The hyperparameters of the model include core parameters, learning control parameters, and metric parameters. Among them, parameters such as learning rate and maximum tree depth directly affect the accuracy of model establishment. Therefore, it is necessary to predetermine the value range of each hyperparameter to be optimized (i.e., the lower and upper bounds of the solution space).

[0163] The hyperparameters to be optimized and the hyperparameter search range are shown below:

[0164]

[0165] S22. A chaotic sequence of random numbers is generated using chaotic mapping, and the chaotic sequence of random numbers is mapped to the hyperparameter search range to obtain the initial noctuid moth population. Each noctuid moth individual in the initial noctuid moth population corresponds to a set of model hyperparameter combinations, and each set of model hyperparameter combinations is the initial position of the corresponding noctuid moth individual.

[0166] In this embodiment of the invention, to improve search capabilities and enhance population diversity, chaotic mapping is introduced to initialize the noctuid moth population, and the steps are as follows:

[0167] Generating Chaotic Sequences: Random chaotic sequences are generated using a sinusoidal chaotic mapping. The calculation formula is as follows:

[0168]

[0169] In the formula, For the first The moth individual was in the first Chaotic sequence values ​​in each hyperparameter dimension For the first The moth individual was in the first Chaotic sequence values ​​in each hyperparameter dimension; A random number between [0, 1] is used to control the generation of chaotic sequences and improve population diversity; ( (Number of individual moths); ( (Number of hyperparameters to be optimized).

[0170] Mapping to Hyperparameter Ranges: Mapping the chaotic sequence to the value ranges of each hyperparameter yields the initial position of each individual moth (i.e., a set of hyperparameter combinations). The calculation formula is as follows:

[0171]

[0172] In the formula, For the first The moth individual was in the first Initial values ​​in each hyperparameter dimension (i.e., a set of hyperparameter combinations corresponding to this moth). , The first The lower and upper bounds of the hyperparameters to be optimized; For the first The moth individual was in the first Chaotic sequence values ​​in each hyperparameter dimension.

[0173] Constructing the moth position matrix: Integrating the initial positions of all individual moths into... OK Column matrix , represented as:

[0174]

[0175] In the formula, This is the position matrix of moths, with dimension 1. Each row of the matrix corresponds to a combination of hyperparameters for an individual moth (i.e., the initial position of the moth).

[0176] S23. Based on the pre-constructed sample set, the initial operating status identification model of the transformer area is trained by using the model hyperparameter combination corresponding to each individual moth, and the model fitness of each individual moth is determined.

[0177] In this embodiment of the invention, a labeled sample set containing four operating states of a transformer substation—no reverse power supply, reverse power supply, islanded operation, and normal operation—is pre-constructed. The sample set is divided into a training set and a validation set according to a preset ratio. The training set for each operating state contains 750 voltage and current data points, and the validation set contains 250 voltage and current data points. Time-frequency multivariate features are extracted from the training set and the validation set. These time-frequency multivariate features are used as the input to the LightGBM initial transformer substation operating state recognition model, and the operating state label is used as the model output. The model is trained using hyperparameter combinations corresponding to each individual moth, and the model's classification performance indicators (such as accuracy, macro-average F1 score, etc.) on the validation set are used as the model fitness for that individual moth.

[0178] To objectively evaluate the model's performance in multi-class tasks, the macro-average principle is used to calculate the model's precision, recall, and F1 score, which serve as the basis for fitness calculation. The steps are as follows:

[0179] For each operating status category of the distribution area, calculate the precision and recall rates separately:

[0180]

[0181] In the formula, For the first Precision of class states For the first Recall rate of class state For the first The number of true positive samples in the class state; For the first The number of false positive samples in each class; For the first The number of false negative samples in the class state; These correspond to four different operating states of the transformer substation.

[0182] Calculate the macro average precision Macro average recall :

[0183]

[0184] In the formula, the arithmetic mean of precision and recall for all categories is taken to obtain the macro average index, which reflects the average performance of the model across all categories.

[0185] Calculate the macro average F1 value :

[0186]

[0187] The calculated macro-average F1 value is used as the model fitness for the individual moth, and is then used for the selection of subsequent light source locations.

[0188] S24. Sort each moth individual in descending order according to the model fitness, and select a preset number of moth individuals from largest to smallest as the initial light source positions.

[0189] In this embodiment of the invention, the model fitness of all individual moths is calculated and sorted in descending order, and the top [number] moths are selected. The position of each individual moth is used as the initial light source position to construct... OK Column light source position matrix , represented as:

[0190]

[0191] In the formula, Here is the light source position matrix, with dimension 1. Each row of the matrix corresponds to the position of a light source (i.e., a set of hyperparameter combinations).

[0192] The number of light sources decreases dynamically with the number of iterations, and the calculation formula is as follows:

[0193]

[0194] In the formula, This represents the number of light sources at the current iteration number. This represents the initial number of light sources; This represents the current iteration number; The maximum number of iterations is preset. This is the rounding function.

[0195] S25. Initialize the flight speed of each individual moth with the hyperparameter search range as a constraint.

[0196] In this embodiment of the invention, to fully utilize the convergence characteristics of moths' flight speed adjustment, the initial flight speed of each moth is randomly initialized, and the calculation formula is as follows:

[0197]

[0198] In the formula, For the first The moth individual was in the first Initial velocity in each hyperparameter dimension; , The first The lower and upper bounds of velocity in each dimension, among which , ; A random number between [0, 1] is used to randomly initialize the speed.

[0199] S26. Calculate the iteration energy based on the current iteration number and the preset maximum iteration number, and compare the iteration energy with the preset energy threshold.

[0200] In this embodiment of the invention, to enhance the matching and connection between exploration and development, and to improve the overall efficiency and stability of the algorithm, iterative energy is introduced. The search strategy is dynamically adjusted, and the calculation formula is:

[0201]

[0202] In the formula, This represents the iteration energy at the current iteration number, used to dynamically switch between global and local search strategies. This represents the current iteration number; The maximum number of iterations is preset; the iteration energy is compared with the preset energy threshold. (Priority 0.4) is compared for subsequent strategy selection.

[0203] S27. When the iteration energy is greater than or equal to the preset energy threshold, the preset global search strategy is used as the target position update strategy. When the iteration energy is less than the preset energy threshold, the preset local optimization strategy is used as the target position update strategy.

[0204] In this embodiment of the invention, a position update strategy is selected based on the comparison result between the iterative energy and the threshold:

[0205] when (In the early stages of iteration, energy is relatively high) A preset global search strategy is adopted, and the position of the moth is updated through a speed mechanism to expand the search area and enhance the exploration capability;

[0206] when (In the later stages of iteration, when the energy is relatively small), a pre-defined local optimization strategy is adopted. The position of the moth is updated by inertial weights to control the flight range and accelerate convergence to the optimal solution.

[0207] S28. Based on the target location update strategy, and combining the initial light source location, the initial location of the moths, and the initial flight speed, update the location of all individual moths to obtain a new moth population.

[0208] Depending on the target location update strategy, the following update process will be performed:

[0209] Furthermore, the preset global search strategy specifically involves shifting and updating the model hyperparameters corresponding to individual moths.

[0210] In this embodiment of the invention, the first calculation is performed. Based on the initial light source position and the initial position of the moth in the next iteration, the Euclidean distance between the individual moth and the light source is calculated to quantify their relative positional relationship:

[0211]

[0212] In the formula, For the first In the nth iteration, the 1st The individual moth and the first The Euclidean distance between light sources is used to measure how close a moth is to a light source. For the first In the nth iteration, the 1st The position of each light source, i.e. the initial light source position, is a combination of hyperparameters of the LightGBM model; For the first In the nth iteration, the 1st The position of each individual moth, i.e. the initial position of the moth, is a set of hyperparameter combinations of the LightGBM model to be optimized; For individual moth indices, the value is... ,in This represents the total number of individual moths. This is the index of the light source, with a value of [value]. ,in This represents the number of light sources in the current iteration. The Euclidean norm operator is used to calculate the distance between two position vectors.

[0213] Then, based on the initialized flight speed, and combining the iterative energy and Euclidean distance, the flight speed of the individual moth is updated:

[0214]

[0215] In the formula, For the first In the nth iteration, the 1st The updated flight speed of an individual moth; For the first In the nth iteration, the 1st The flight speed of an individual moth, i.e. the initial flight speed, is the speed value that was initially randomly generated or updated in the last iteration; Here, r is a random coefficient used to control the search path of moths, with a value range of [r, 1], where r is dynamically determined by the iterative energy and the threshold:

[0216]

[0217] In the formula, This is a dynamic adjustment coefficient used to control the search path of moths. A preset energy threshold is used to divide the global search and local optimization stages.

[0218] Finally, combining the initial light source position, Euclidean distance (calculated from the moth's initial position), and updated flight speed (obtained from the initialized flight speed), the position of the individual moth is updated using the following formula:

[0219]

[0220] In the formula, For the first In the nth iteration, the 1st The updated position of each individual moth, i.e., a new set of LightGBM model hyperparameters. The preset flight adjustment coefficient is used to control the spiral trajectory of the moth flying around the light source.

[0221] Furthermore, the pre-defined local optimization strategy specifically involves fine-tuning and updating the hyperparameters of the model corresponding to each individual moth.

[0222] In this embodiment of the invention, firstly, with the first... Based on the initial light source position and the initial position of the moth in the next iteration, the Euclidean distance between the individual moth and the light source is calculated to quantify their relative positional relationship, providing basic parameters for subsequent position updates. The formula is as follows:

[0223]

[0224] In the formula, For the first In the nth iteration, the 1st The individual moth and the first The Euclidean distance between light sources is used to measure how close a moth is to a light source. For the first In the nth iteration, the 1st The position of each light source, i.e. the initial light source position, is a combination of hyperparameters of the LightGBM model; For the first In the nth iteration, the 1st The position of each individual moth, i.e. the initial position of the moth, is a set of hyperparameter combinations of the LightGBM model to be optimized; For individual moth indices, the value is... ,in This represents the total number of individual moths. This is the index of the light source, with a value of [value]. ,in This represents the number of light sources in the current iteration. The Euclidean norm operator is used to calculate the distance between two position vectors.

[0225] Then, to control the flight range of the moth and avoid oscillations in the later stages of the iteration, an inertial weight is introduced. This weight is dynamically calculated from the energy of the current iteration and is used to adjust the amplitude of the position update. The formula is as follows:

[0226]

[0227] In the formula, For the first The inertia weight in the next iteration is used to control the magnitude of the moth's position update. The smaller the weight, the smaller the magnitude of the position update, and the higher the local search accuracy of the algorithm.

[0228] Finally, combining the initial light source position, Euclidean distance (calculated from the moth's initial position), updated flight speed (obtained from the initialized flight speed), and calculated inertial weights, the position of the individual moth is fine-tuned and updated using the following formula:

[0229]

[0230] In the formula: For the first In the nth iteration, the 1st The updated position of each individual moth, i.e., a new set of LightGBM model hyperparameters; For the first In the nth iteration, the 1st The updated flight speed of each individual moth is based on its initial flight speed. The velocity update formula in the global search strategy is used to calculate the velocity, which provides a velocity driver for the position update of moths.

[0231] Through the above steps, the local optimization strategy, combined with three core elements, completed the fine-tuning and updating of the moth's position, achieving local convergence in the later stages of iteration. Together with the global search strategy, it constitutes a complete hyperparameter optimization process for the moth phototaxis algorithm.

[0232] S29. Based on the new noctuid moth population, jump to the steps based on the pre-divided training sample set until the current iteration number equals the preset maximum iteration number. Then, select the set of model hyperparameters with the highest current model fitness as the optimal model hyperparameter combination.

[0233] In this embodiment of the invention, after the position update is completed, based on the new noctuid moth population, the process jumps to step S23 to recalculate the model fitness of each individual noctuid moth; the noctuid moth fitness of this iteration is merged and sorted with the light source fitness of the previous iteration, and the top fitness values ​​are selected. Update the light source position and the iteration count for larger locations. If the current iteration number Equal to the preset maximum number of iterations If the model fitness is not found, the iteration terminates and the hyperparameter combination with the highest fitness is selected as the optimal model hyperparameter combination; otherwise, the next iteration continues.

[0234] To visually demonstrate the iterative optimization process of the phototaxis algorithm for moths, the error variation curve of the algorithm iteration process in this embodiment is shown as follows: Figure 3 As shown in the figure, the horizontal axis represents the number of iterations, and the vertical axis represents the model training error. From the iteration process curve, it can be seen that as the number of iterations increases, the model error shows a trend of rapid decrease and gradual convergence: in the early stage of iteration (0~100 times), the error decreases rapidly, the algorithm is in the global exploration stage, and the hyperparameter search range is rapidly narrowed; in the middle and late stages of iteration (after 100 times), the error decrease tends to be gradual and eventually stabilizes, the algorithm enters the local optimization stage, and completes the convergence of the optimal solution. This shows that the moth phototaxis algorithm has good convergence performance, can effectively complete hyperparameter optimization, and avoid getting trapped in local optima.

[0235] After the iteration terminates, the optimal hyperparameter combination of the LightGBM model is shown in the table below:

[0236]

[0237] The LightGBM model with the optimal hyperparameter combination has the best classification performance in the task of identifying the operating status of transformer substations, and can be directly substituted into the initial model to construct the target transformer substation operating status identification model.

[0238] S210. Substitute the optimal model hyperparameter combination into the initial transformer area operation status identification model to obtain the target transformer area operation status identification model.

[0239] In this embodiment of the invention, the optimal model hyperparameter combination is substituted into the initial transformer area operation status identification model to complete the model parameter optimization and obtain the target transformer area operation status identification model.

[0240] After completing the model construction and parameter optimization in step 205 to obtain the target area operation status identification model, in order to verify the model's generalization ability and actual identification accuracy, this embodiment uses an independent test set to conduct multi-class performance evaluation of the model. The specific process is as follows:

[0241] Dataset preparation and partitioning:

[0242] A pre-constructed dataset of labeled three-phase electrical parameters is built, including four operating states of the transformer substation: no reverse power supply, reverse power supply, islanded operation, and normal operation. The dataset is divided into a training set and a test set in a 3:1 ratio. The training set for each operating state contains 750 voltage and current data points, and the test set contains 250 voltage and current data points. The training set has been used for model training and hyperparameter optimization in step 205, while the test set is independent data that the model has not encountered. It is used to objectively evaluate the generalization performance of the model and avoid performance evaluation bias caused by data leakage.

[0243] Test set feature extraction:

[0244] For each three-phase electrical parameter data in the test set, the process is carried out according to steps 201-204: First, the α component and zero-sequence component of the three-phase electrical parameter are obtained through Clarke transform, and then the complex time-frequency matrix is ​​obtained through polynomial adaptive discrete S-transform. Subsequently, the modulus of the complex time-frequency matrix is ​​taken to obtain the modulus time-frequency matrix, and combined with the time-time transformation matrix, multi-dimensional features in the time domain, frequency domain, and time-time domain are extracted to finally obtain the time-frequency multi-dimensional features of the test set. This feature extraction process is completely consistent with the feature processing process of the training set, ensuring the consistency of data processing and avoiding performance evaluation distortion caused by differences in feature processing.

[0245] Model testing and confusion matrix analysis:

[0246] The target transformer area operation status identification model, after optimizing the input parameters of the time-frequency multivariate features of the test set, outputs the predicted operation status of each test sample. The predicted results are then compared with the actual status labels to obtain the model's classification confusion matrix, such as... Figure 4 As shown: The rows of the confusion matrix correspond to the actual values ​​of the operating status of the transformer area, and the columns correspond to the predicted values ​​of the model. The diagonal elements of the matrix represent the number of samples correctly identified for each status, and the off-diagonal elements represent the number of samples misclassified. From the confusion matrix, we can see that:

[0247] Of the 250 test samples without reverse power supply, 240 were correctly identified, while only 3, 3, and 4 were misidentified as reverse power supply, islanded operation, and normal operation, respectively.

[0248] Of the 250 test samples in the reverse power supply state, 236 were correctly identified, while a small number of samples were misidentified as other states.

[0249] Of the 250 test samples in the isolated operation state, 236 were correctly identified;

[0250] Of the 250 test samples in normal operation, 243 were correctly identified. Overall, the model has a high accuracy rate in identifying the operating status of various transformer substations, with only a small number of cross-status misclassifications, indicating that the model has stable classification performance and strong ability to distinguish between different operating statuses.

[0251] Comparison of performance index calculation and optimization effects:

[0252] Using the same macro-average principle as in step S23 for model fitness calculation, the model's precision, recall, and F1 score were calculated to evaluate its overall performance in multi-class tasks. Simultaneously, the model was compared with the original LightGBM model without optimization using the moth phototaxis algorithm to verify the effectiveness of hyperparameter optimization.

[0253] Performance of the target station operation status identification model after parameter optimization: macro average precision is 0.9551, macro average recall is 0.9550, and macro average F1 score is 0.9551;

[0254] Performance of the original LightGBM model without parameter optimization: macro-mean precision is 0.9129, macro-mean recall is 0.9120, and macro-mean F1 score is 0.9125. The comparison results show that after hyperparameter optimization using the moth phototaxis algorithm, the model's macro-mean precision, recall, and F1 score are significantly improved. Specifically, the macro-mean F1 score, which comprehensively reflects classification performance, increased from 0.9125 to 0.9551, an improvement of approximately 4.26 percentage points. This indicates that parameter optimization effectively solved the overfitting or underfitting problems of the original model, improved the model's generalization ability and recognition accuracy, and provided reliable performance assurance for subsequent identification of transformer station operation status in real-world scenarios.

[0255] Hyperparameter search range: The upper and lower limits of the preset values ​​for each hyperparameter to be optimized in the initial transformer area operation status identification model; Chaotic mapping: A mapping function that generates a sequence with pseudo-random ergodic characteristics for population initialization; Random number chaotic sequence: A discrete numerical sequence with uniform distribution and ergodicity generated by chaotic mapping; Initial moth population: The set of all moth individuals formed after mapping the chaotic sequence to the hyperparameter search range; Moth individual: An independent optimization unit in the moth phototaxis algorithm, each individual corresponding to a set of model hyperparameter combinations; Model hyperparameter combination: A set of parameter configurations composed of multiple hyperparameters to be optimized in the model; Initial moth position: The model hyperparameter combination corresponding to each moth individual in the initial moth population; Sample set: A pre-constructed electrical parameter dataset with real labels of transformer area operation status for model training and fitness evaluation; Model fitness: A numerical value used to quantitatively evaluate the quality of the hyperparameter combination corresponding to the moth individual, with model classification performance as the evaluation index; Initial light source position: The high-fitness moth individuals selected after sorting them in descending order of model fitness. Individual position; Flight speed: A variable representing the step size of the position update of each moth individual in each iteration of the phototaxis algorithm; Iteration energy: A control parameter used to switch optimization strategies, calculated from the current iteration number and the preset maximum iteration number; Preset energy threshold: A manually set critical value for iteration energy, used to divide the global search and local optimization stages; Target position update strategy: The global search strategy or local optimization strategy selected based on the comparison result of iteration energy and preset energy threshold; Preset global search strategy: An optimization strategy that performs a large-scale update of the position of moth individuals when the iteration energy meets the threshold condition; Preset local optimization strategy: An optimization strategy that performs a small-scale fine update of the position of moth individuals when the iteration energy is lower than the threshold condition; New moth population: A new generation population composed of all moth individuals after position update; Preset maximum iteration number: The maximum number of iterations allowed by the phototaxis algorithm, used as the algorithm termination condition; Optimal model hyperparameter combination: The set of model hyperparameters with the highest fitness value after the algorithm iteration terminates.

[0256] Offset Update: Under the global search strategy, this method updates the hyperparameters of the model corresponding to each individual moth over a large range by combining the light source position, the initial position of the moth, and its flight speed.

[0257] Fine-tuning update: Under the local optimization strategy, an inertial weight is introduced to constrain the update range, and the hyperparameters of the model corresponding to the individual moth are finely modified and updated in a small range.

[0258] Step 206: Input the time-frequency multi-dimensional features into the target area operation status identification model to perform status identification and obtain the operation status identification result of the area.

[0259] In this embodiment of the invention, the time-frequency multi-dimensional features obtained by processing the data to be identified are input into the target area operation status identification model optimized by the moth phototaxis algorithm. The model automatically analyzes and classifies the input features and directly outputs the operation status identification result corresponding to the current distributed photovoltaic area.

[0260] This invention has the following advantages:

[0261] Adaptive variable resolution time-frequency analysis to extract high-discrimination signal features

[0262] This scheme is based on the Clarke transform components (α component and zero-sequence component) obtained from the three-phase electrical parameters of the distributed photovoltaic power station through Clarke transform. It adopts a polynomial adaptive discrete S-transform and combines it with a Gaussian window function optimized for the power station operation status identification task. This Gaussian window function has adaptive variable resolution characteristics through preset window function control adjustment parameters: for power station operation status such as reverse power supply, it can provide high time resolution and accurately capture its disturbance characteristics; for power station operation status such as no reverse power supply, islanded operation, and normal operation, it has better frequency resolution detection accuracy, effectively identifies its disturbance frequency, and finally extracts signal features with obvious discrimination, providing a reliable foundation for subsequent status identification.

[0263] Multi-dimensional time-frequency multi-dimensional feature extraction comprehensively characterizes the operating status of the transformer area.

[0264] This scheme extracts the time-domain and frequency-domain features of the transformer operating status based on the modulo time-frequency matrix obtained by taking the modulus of the complex time-frequency matrix; at the same time, it extracts the time-time domain features based on the time-time transform matrix obtained by the discrete inverse Fourier transform; the three types of features are fused to form a time-frequency multi-dimensional feature, which can effectively capture the signal mode, energy characteristics and time series change trends in the transformer operating signal. The entire extraction process does not require setting a threshold, providing a comprehensive and reliable feature basis for the accurate identification of the transformer operating status.

[0265] The LightGBM model, optimized with the phototaxis algorithm for moths, achieves high-precision and high-efficiency identification.

[0266] This scheme employs the phototaxis algorithm of moths to optimize the hyperparameters of the Lightweight Gradient Boosting Machine (LightGBM) model for identifying the operating status of transformer substations. This results in a non-invasive identification method that does not require threshold setting, improving computational efficiency while maintaining accuracy. The algorithm introduces chaotic mapping to initialize the moth population, enhancing population ergonomics and laying the foundation for global search. An innovative iterative energy switching strategy dynamically switches between global search and local optimization strategies based on the search progress, effectively balancing the algorithm's global exploration and local development capabilities. Furthermore, the combination of a velocity update mechanism and an inertial weight update method improves the accuracy of hyperparameter search. These three elements work synergistically to significantly enhance the classification accuracy of the LightGBM model in identifying the operating status of transformer substations. Testing verifies that the optimized model's macro-average F1 score is approximately 4.26 percentage points higher than the unoptimized model, demonstrating a significant improvement in recognition performance.

[0267] This invention provides a distributed photovoltaic (PV) transformer area operation status identification scheme that requires neither injecting disturbance signals into the system nor manually setting identification thresholds. It collects three-phase electrical parameters at the common coupling point of the distributed PV system, performs Clarke transform, and then uses a polynomial adaptive S-transform to obtain a complex time-frequency matrix. This matrix is ​​then further processed by a discrete inverse Fourier transform to obtain a time-to-time transform matrix. Based on these matrices, time-domain, frequency-domain, and time-domain multi-dimensional features are extracted. An initial transformer area operation status identification model is then constructed. The model's hyperparameters are optimized using a moth phototaxis algorithm that incorporates chaotic mapping to initialize the population and iterative energy switching strategies to obtain a target transformer area operation status identification model. Finally, the time-frequency multi-dimensional features are input into the target model to achieve status identification. The polynomial adaptive S-transform optimizes window function parameters for complex operating conditions such as reverse power supply and islanded operation, achieving adaptive time-frequency resolution. This provides high time resolution for disturbance signals to capture instantaneous changes and high frequency resolution for steady-state signals to detect frequency features. The extraction of highly discriminative time-frequency features, along with multi-dimensional feature extraction in the time, frequency, and time domains, effectively captures the patterns, energy, and trends of the operating signals of the distribution area. It comprehensively characterizes signal differences under different operating conditions without relying on threshold settings, avoiding the subjectivity and blind spots of traditional threshold methods. In the moth phototaxis algorithm, the initial moth position generated based on chaotic mapping ensures the traversal of the initial population, laying the foundation for global search. The initial light source position, serving as a high-quality solution reference, provides stable directional guidance for hyperparameter optimization. Combined with the dynamically adjusted flight speed using an iterative energy switching strategy, it achieves an effective balance between global exploration and local development. The synergistic effect of these three elements significantly improves the optimization effect of the model's hyperparameters, enabling the optimized target model to fully learn multi-feature patterns under complex operating conditions. Ultimately, this greatly improves the accuracy of identifying the operating status of distributed photovoltaic distribution areas under complex conditions such as reverse power transmission and islanded operation, effectively solving the problem of insufficient identification efficiency under complex conditions in existing technologies.

[0268] Please see Figure 5 , Figure 5 This is a structural block diagram of a distributed photovoltaic power station operation status identification system provided in Embodiment 3 of the present invention.

[0269] This invention provides a distributed photovoltaic (PV) transformer area operation status identification system, comprising:

[0270] The conversion module 501 is used to collect the three-phase electrical parameters at the common coupling point of the distributed photovoltaic system and perform Clarke transformation to obtain Clarke transformation components.

[0271] Analysis module 502 is used to determine the time-to-time transformation matrix based on the complex time-frequency matrix obtained by adaptive S-transform of the Clark transform components;

[0272] Extraction module 503 is used to perform multivariate feature extraction using complex time-frequency matrix and time-time transformation matrix to obtain time-frequency multivariate features;

[0273] Optimization module 504 is used to construct an initial transformer area operation status identification model and optimize the parameters of the initial transformer area operation status identification model through the moth phototaxis algorithm to obtain the target transformer area operation status identification model.

[0274] The identification module 505 is used to input the time-frequency multi-dimensional features into the target area operation status identification model for status identification and obtain the operation status identification result of the area.

[0275] Since the above is a system corresponding to a method for identifying the operating status of a distributed photovoltaic power station, and its implementation principle is the same as that of a method for identifying the operating status of a distributed photovoltaic power station, for the sake of convenience and brevity, those skilled in the art can clearly understand that the specific working process of the system and modules described above can be referred to the corresponding process in the aforementioned method embodiments, and will not be repeated here.

[0276] Please see Figure 6 , Figure 6 This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention.

[0277] An electronic device according to an embodiment of the present invention includes: a memory 601 and a processor 602. The memory 601 stores a computer program. When the computer program is executed by the processor 602, the processor 602 executes the distributed photovoltaic area operation status identification method as described in the above embodiment.

[0278] Memory 601 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Memory 601 has storage space 603 for program code 613 for performing any of the method steps described above. For example, storage space 603 for program code may include various program codes 613 for implementing the various steps in the methods described above. This program code may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When run by a computing processing device, this code causes the computing processing device to perform the various steps in the methods described above. This program code may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, CDs, memory cards, or floppy disks. The program code may be compressed, for example, in a suitable form. When these codes are run by a computing device, the computing device causes the device to perform the various steps in the distributed photovoltaic area operation status identification method described above.

[0279] Embodiment 5 of the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the distributed photovoltaic area operation status identification method as described in the above embodiments.

[0280] Embodiment 6 of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer performs the distributed photovoltaic area operation status identification method as described in the above embodiments.

[0281] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0282] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0283] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0284] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0285] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0286] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A distributed photovoltaic substation operation state recognition method, characterized in that, include: The three-phase electrical parameters at the common coupling point of the distributed photovoltaic system are collected and subjected to Clarke transformation to obtain the Clarke transformation components. The time-frequency transformation matrix is ​​determined based on the complex time-frequency matrix obtained by adaptive S-transform of the Clarke transform components. Multivariate feature extraction is performed using the complex time-frequency matrix and the time-time transformation matrix to obtain time-frequency multivariate features; An initial transformer area operation status identification model is constructed, and the parameters of the initial transformer area operation status identification model are optimized by the noctuid moth phototaxis algorithm to obtain the target transformer area operation status identification model. The time-frequency multivariate features are input into the target transformer area operation status identification model for status identification, and the transformer area operation status identification result is obtained.

2. The distributed photovoltaic regional operation state recognition method according to claim 1, characterized in that, The step of determining the time-frequency transformation matrix based on the complex time-frequency matrix obtained by adaptive S-transform of the Clarke transform components includes: Based on the Gaussian window function, a polynomial adaptive S-transform is performed on the Clarke transform components to obtain a complex time-frequency matrix; Perform a discrete inverse Fourier transform on the complex time-frequency matrix to obtain the time-time transform matrix.

3. The distributed photovoltaic regional operation state recognition method according to claim 1, characterized in that, The time-frequency multivariate features include time-domain features, frequency-domain features, and time-time-domain features. The extraction of multivariate features using the complex time-frequency matrix and the time-time transform matrix to obtain the time-frequency multivariate features includes: Taking the modulus of the complex time-frequency matrix yields the modulus time-frequency matrix; The time-domain features are obtained by using the modulus-time-frequency matrix. Frequency domain features are obtained by using the modulus-time-frequency matrix; The time-domain features are obtained by using the time-time transformation matrix.

4. The distributed photovoltaic regional operation state recognition method according to claim 1, characterized in that, The process of constructing an initial transformer area operation status identification model and optimizing the parameters of the initial transformer area operation status identification model using a moth phototaxis algorithm to obtain a target transformer area operation status identification model includes: Construct an initial transformer area operation status identification model and determine the hyperparameter search range; A chaotic sequence of random numbers is generated using chaotic mapping, and the chaotic sequence of random numbers is mapped to the hyperparameter search range to obtain the initial noctuid moth population; Each noctuid moth in the initial noctuid moth population corresponds to a set of model hyperparameter combinations, and each set of model hyperparameter combinations represents the initial position of the noctuid moth for the corresponding noctuid moth individual. Based on a pre-constructed sample set, the initial operating status identification model of the transformer area is trained using the model hyperparameter combination corresponding to each individual moth, and the model fitness of each individual moth is determined. Based on the model fitness, each moth individual is sorted in descending order, and a preset number of moth individuals are selected from largest to smallest as the initial light source positions. The flight speed of each individual moth is initialized using the hyperparameter search range as a constraint. The iteration energy is calculated based on the current iteration number and the preset maximum iteration number, and then compared with the iteration energy and the preset energy threshold. When the iteration energy is greater than or equal to the preset energy threshold, the preset global search strategy is used as the target position update strategy; when the iteration energy is less than the preset energy threshold, the preset local optimization strategy is used as the target position update strategy. Based on the target position update strategy, and combining the initial light source position, the initial position of the moth, and the initialized flight speed, the positions of all individual moths are updated to obtain a new moth population. Based on the new noctuid moth population, the process jumps to the step based on the pre-divided training sample set until the current iteration number equals the preset maximum iteration number. Then, the set of model hyperparameters with the highest current model fitness is selected as the optimal model hyperparameters combination. Substituting the optimal model hyperparameter combination into the initial transformer area operation status identification model yields the target transformer area operation status identification model.

5. The distributed photovoltaic station operation state recognition method according to claim 4, characterized in that, The preset global search strategy specifically involves shifting and updating the hyperparameters of the model corresponding to individual moths.

6. The distributed photovoltaic station operation state recognition method according to claim 4, characterized in that, The preset local optimization strategy specifically involves fine-tuning and updating the hyperparameters of the model corresponding to individual moths.

7. A distributed photovoltaic substation operation state identification system, characterized in that, include: The conversion module is used to collect the three-phase electrical parameters at the common coupling point of the distributed photovoltaic system and perform Clarke transformation to obtain Clarke transformation components. The analysis module is used to determine the time-to-time transformation matrix based on the complex time-frequency matrix obtained by adaptive S-transform of the Clarke transform components; The extraction module is used to perform multivariate feature extraction using the complex time-frequency matrix and the time-time transformation matrix to obtain time-frequency multivariate features; The optimization module is used to construct an initial transformer area operation status identification model and optimize the parameters of the initial transformer area operation status identification model using the noctuid moth phototaxis algorithm to obtain the target transformer area operation status identification model. The identification module is used to input the time-frequency multi-dimensional features into the target area operation status identification model for status identification, and obtain the operation status identification result of the area.

8. An electronic device, comprising: The system includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the steps of the distributed photovoltaic power station operation status identification method as described in any one of claims 1-6.

9. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed, it implements the distributed photovoltaic power station operation status identification method as described in any one of claims 1-6.

10. A computer program product, characterised in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, wherein when the program instructions are executed by a computer, the computer performs the distributed photovoltaic area operation status identification method as described in any one of claims 1-6.

Images