Power grid alarm sensing system, method and equipment based on artificial intelligence and medium

Abstract

The invention discloses a power grid alarm sensing system, method and device based on artificial intelligence and a medium, and relates to the technical field of computers, and the system comprises a power grid alarm sensing middle station, a data acquisition module, a feature extraction module, a model matching module, a model prediction module and a current imbalance alarm sensing module. The data acquisition module is used for acquiring three-phase current to obtain current time sequence data; a feature extraction module performs feature extraction on the current time sequence data to obtain multi-dimensional feature extraction; a model matching module performs matching in a preset model library based on the electrical physical characteristics to obtain a target current imbalance sensing model; the model prediction module inputs the electrical operation state characteristics into a target current unbalance perception model to obtain a current unbalance perception result; and the current imbalance alarm sensing module carries out current imbalance alarm based on the current imbalance sensing result. According to the invention, the working condition of the power grid is described more comprehensively, and current imbalance alarm sensing is accurately carried out on the power grid.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an artificial intelligence-based power grid alarm perception system, method, device and medium. Background Art

[0002] In existing power grid operation monitoring, current imbalance alarm perception is an important link to ensure the safe and stable operation of the power grid. Traditional power grid current imbalance alarm perception methods are mainly based on electrical quantity measurement algorithms. The common method is to measure the amplitude and phase of the three-phase current, and then determine whether there is a current imbalance phenomenon based on the difference between the three-phase currents or the imbalance calculation formula (such as the ratio of negative sequence current to positive sequence current). However, in modern power grids, there are a large number of nonlinear loads, distributed power supply access and the widespread use of power electronic equipment. These factors will cause harmonic pollution and current distortion in the power grid, making the amplitude and phase information of the traditionally measured three-phase current inaccurate, thereby causing errors in the current imbalance calculation based on this, and failing to accurately perform current imbalance alarm perception on the power grid. Summary of the Invention

[0003] In view of the above-mentioned problems, the present invention is proposed.

[0004] Therefore, the technical problem solved by the present invention is: how to propose an artificial intelligence-based power grid alarm perception system, method, equipment and medium to achieve a more comprehensive description of the complex working conditions of the power grid and accurately perceive the current imbalance alarm of the power grid.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: an artificial intelligence-based power grid alarm perception system, comprising: a power grid alarm perception center, a data acquisition module, a feature extraction module, a model matching module, a model prediction module and a current imbalance alarm perception module; the power grid alarm perception center is respectively connected to the data acquisition module, the feature extraction module, the model matching module, the model prediction module and the current imbalance alarm perception module to manage each module; the data acquisition module is used to collect the three-phase current in the power grid based on distributed sensors to obtain current time series data; the feature extraction module is used to analyze the current time series data. The column data is subjected to feature extraction to obtain multi-dimensional feature extraction; the multi-dimensional feature extraction includes electrical physical features and electrical operation status features; a model matching module is used to match in a preset model library based on the electrical physical features to obtain a target current imbalance perception model; the target current imbalance perception model is trained based on sample features and perception result labels; a model prediction module is used to input the electrical operation status features into the target current imbalance perception model to obtain a current imbalance perception result output by the target current imbalance perception model; a current imbalance alarm perception module is used to perform a current imbalance alarm on the power grid based on the current imbalance perception result.

[0006] Another object of the present invention is to provide an artificial intelligence-based power grid alarm perception method.

[0007] To solve the above technical problems, the present invention provides the following technical solution: a power grid alarm perception method based on artificial intelligence, which includes the following steps:

[0008] Based on distributed sensors, the three-phase current in the power grid is collected to obtain current time series data; feature extraction is performed on the current time series data to obtain multi-dimensional feature extraction; the multi-dimensional feature extraction includes electrical physical features and electrical operating status features; based on the electrical physical features, a target current imbalance perception model is obtained by matching in a preset model library; the target current imbalance perception model is trained based on sample features and perception result labels; the electrical operating status features are input into the target current imbalance perception model to obtain the current imbalance perception results output by the target current imbalance perception model; based on the current imbalance perception results, a current imbalance alarm is issued to the power grid.

[0009] As a preferred solution of the artificial intelligence-based power grid alarm perception method described in the present invention, wherein: the electrical physical characteristics include three-phase current spectrum characteristics, three-phase current transient characteristics and three-phase current high-frequency component characteristics; based on the electrical physical characteristics, matching in a preset model library to obtain a target current imbalance perception model, including: based on the three-phase current spectrum characteristics, the three-phase current transient characteristics and the three-phase current high-frequency component characteristics, determining the comprehensive matching value of each current imbalance perception model in the preset model library to the electrical physical characteristics of the power grid; traversing the comprehensive matching value of each current imbalance perception model to obtain the maximum comprehensive matching value; determining the current imbalance perception model corresponding to the maximum comprehensive matching value as the target current imbalance perception model.

[0010] As a preferred solution of the artificial intelligence-based power grid alarm perception method described in the present invention, wherein: based on the three-phase current spectrum characteristics, the three-phase current transient characteristics and the three-phase current high-frequency component characteristics, the comprehensive matching value of each current imbalance perception model in the preset model library to the electrical physical characteristics of the power grid is determined, including: based on the three-phase current spectrum characteristics and the optimal spectrum characteristics of each current imbalance perception model for the three-phase current, determining the similarity of the spectrum characteristics of each current imbalance perception model; based on the three-phase current transient characteristics and the optimal transient characteristics of each current imbalance perception model for the three-phase current, determining the current transient matching value of each current imbalance perception model; based on the three-phase current high-frequency component characteristics and the optimal adaptive high-frequency characteristics of each current imbalance perception model for the three-phase current, determining the high-frequency feature adaptation value of each current imbalance perception model; based on the spectrum characteristic similarity, current transient matching value and high-frequency feature adaptation value of each current imbalance perception model.

[0011] As a preferred solution of the artificial intelligence-based power grid alarm perception method described in the present invention, wherein: the electrical operation status characteristics include three-phase current amplitude, three-phase current phase and three-phase current harmonic content; the target current imbalance perception model includes a data processing layer and a model prediction layer; the electrical operation status characteristics are input into the target current imbalance perception model to obtain the current imbalance perception result output by the target current imbalance perception model, including: inputting the three-phase current amplitude, the three-phase current phase and the three-phase current harmonic content into the target current imbalance perception model, for the data processing layer, generating an amplitude matrix according to the three-phase current amplitude and the three-phase current phase, and generating a harmonic matrix according to the three-phase current harmonic content; for the model prediction layer, fusing the amplitude matrix and the harmonic matrix to obtain a fused tensor, and transforming the fused tensor to obtain a transformed result; current imbalance perception is performed based on the transformed result, the fused tensor and the three-phase current amplitude, and the current imbalance perception result is output.

[0012] As a preferred solution of the artificial intelligence-based power grid alarm perception method described in the present invention, wherein: current imbalance perception is performed based on the transformed result, the fused tensor and the three-phase current amplitude, and the current imbalance perception result is output, including: determining the current imbalance probability based on the transformed result; the current imbalance probability characterizes the possibility that the power grid is in a current imbalance state; determining the alarm level coefficient of current imbalance based on the transformed result, the fused tensor and the three-phase current amplitude; determining the current imbalance perception result based on the current imbalance probability and the alarm level coefficient.

[0013] As a preferred solution of the artificial intelligence-based power grid alarm perception method described in the present invention, wherein: the training constraints are: all loss function values ​​are less than or equal to a preset loss threshold, the loss difference between two adjacent loss function values ​​is less than or equal to a preset difference, and the learning rate in the target optimization function is within a preset range, the target optimization function is constructed based on the model parameters of the deep learning network, and the model parameters include weight parameters and bias term parameters; training the target current imbalance perception model includes the following steps: inputting sample features into the deep learning network, obtaining the perception prediction results of each sample feature output by the deep learning network, and determining the loss function value of each sample feature based on the loss function of the deep learning network and the perception prediction results and perception result labels of each sample feature; if there is When at least one loss function value is greater than a preset loss threshold, or / and there is at least one loss difference greater than a preset difference, or / and the learning rate of the deep learning network is outside a preset range, the model update coefficient is determined based on the perception prediction result and the perception result label of each sample feature; the learning rate and function parameters of the target optimization function are updated based on the model update coefficient to obtain an updated deep learning network; the sample features are input into the updated deep learning network to obtain the perception prediction result of each sample feature output by the updated deep learning network, until the loss function value obtained based on the loss function and the perception prediction result and the perception result label of each sample feature, and the learning rate of the updated deep learning network meet the training constraints, and the target current imbalance perception model is obtained.

[0014] As a preferred solution of the artificial intelligence-based power grid alarm perception method of the present invention, the specific formula of the loss function is as follows:

[0015]

[0016] in, is the loss function value of the i-th sample feature; y i is the perception result label of the i-th sample feature; is the perceptual prediction result of the i-th sample feature; the specific formula of the target optimization function is as follows:

[0017]

[0018] Among them, θ t+1 is the model parameter of the t+1th iteration; θ t is the model parameter of the tth iteration; η t is the learning rate of the tth iteration; m t is the momentum term of the t-th iteration; δ is the preset adjustment parameter; is the gradient of the loss function at the tth iteration; J(θ) is the loss function; is the loss function value of the sample feature; θ is the parameter vector of the model; θ k is the kth parameter vector; λ is the regularization parameter; |θ k | 1.5 is the regularization term.

[0019] The present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the processor executes the computer program, the steps of the artificial intelligence-based power grid alarm perception system are implemented.

[0020] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, the steps of the artificial intelligence-based power grid alarm perception system are implemented.

[0021] Beneficial effects of the present invention: The artificial intelligence-based power grid alarm perception system provided by the present invention extracts features from the current time series data obtained by three-phase current collection to obtain the electrical physical characteristics and electrical operating status characteristics of the three-phase current, and performs current imbalance perception through the electrical physical characteristics and electrical operating status characteristics, which can more comprehensively describe the complex working conditions of the power grid; on the other hand, a target current imbalance perception model that is best suitable for the current electrical physical characteristics is matched in the preset model library, and then the current imbalance perception is performed on the current electrical operating status characteristics through the target current imbalance perception model to obtain accurate current imbalance perception results, thereby accurately performing current imbalance alarm perception on the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 This is a structural diagram of the artificial intelligence-based power grid alarm perception system in Example 1.

[0024] Figure 2 This is a flow chart of the artificial intelligence-based power grid alarm perception method in Example 2.

[0025] Figure 3 This is an example diagram of the electronic device in Example 3.

[0026] Figure 4 This is an example diagram of the computer-readable storage medium in Example 3. DETAILED DESCRIPTION

[0027] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0028] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0029] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0030] Example 1, with reference to Figure 1 , which is the first embodiment of the present invention, provides an artificial intelligence-based power grid alarm perception system.

[0031] The existing grid current imbalance alarm perception methods mainly have the following problems: the common method is to measure the amplitude and phase of the three-phase current, and then determine whether there is a current imbalance phenomenon based on the difference of the three-phase current or the imbalance calculation formula (such as the ratio of negative sequence current to positive sequence current); however, in modern power grids, there are a large number of nonlinear loads, distributed power supply access and the widespread use of power electronic equipment. These factors will cause harmonic pollution and current distortion in the power grid, making the amplitude and phase information of the traditionally measured three-phase current inaccurate, thereby causing errors in the current imbalance calculation based on this, and it is impossible to accurately perceive the current imbalance alarm of the power grid.

[0032] The present invention provides an effective solution to the above-mentioned problems. Next, we will combine multiple embodiments to elaborate on how to implement the artificial intelligence-based power grid alarm perception system.

[0033] Figure 1 The schematic diagram of the structure of the power grid alarm perception system based on artificial intelligence is shown, including:

[0034] Power grid alarm perception center, data acquisition module, feature extraction module, model matching module, model prediction module and current imbalance alarm perception module.

[0035] In an embodiment of the present invention, the power grid alarm perception center is connected to the data acquisition module, feature extraction module, model matching module, model prediction module and current imbalance alarm perception module respectively, so that each module can be managed.

[0036] In the embodiment of the present invention, the data acquisition module includes distributed sensors. Therefore, the data acquisition module calls the distributed sensors to collect three-phase currents in the power grid to obtain current time series data.

[0037] In the embodiment of the present invention, the feature extraction module performs feature extraction on the current time series data to obtain multi-dimensional feature extraction.

[0038] Furthermore, multi-dimensional feature extraction includes electrical physical features and electrical operating status features.

[0039] Furthermore, the electrical physical characteristics include three-phase current spectrum characteristics, three-phase current transient characteristics and three-phase current high-frequency component characteristics.

[0040] Furthermore, the electrical operation status characteristics include three-phase current amplitude, three-phase current phase and three-phase current harmonic content.

[0041] In this embodiment of the present invention, a preset model library is pre-established, comprising multiple current imbalance perception models. Each current imbalance perception model has different optimal applicable parameters, and each current imbalance perception model is trained based on sample features and perception result labels. Therefore, the model matching module matches the preset model library based on the electrical and physical characteristics, obtaining the target current imbalance perception model from the preset model library that best matches the current electrical and physical characteristics. Specifically, the target current imbalance perception model from the preset model library that best matches the current three-phase current spectrum characteristics, three-phase current transient characteristics, and three-phase current high-frequency component characteristics is matched.

[0042] In an embodiment of the present invention, the model prediction module inputs the current three-phase current amplitude, three-phase current phase and three-phase current harmonic content into the target current imbalance perception model. The target current imbalance perception model performs current imbalance perception prediction based on the electrical operating state characteristics and outputs the current imbalance perception result.

[0043] In this embodiment of the present invention, the current imbalance sensing result includes a current imbalance probability and a current imbalance alarm level coefficient. The current imbalance probability represents the likelihood that the power grid is in a current imbalance state. Therefore, when the current imbalance alarm sensing module determines, based on the current imbalance sensing result, that the current imbalance probability is greater than a preset trigger probability, and that the power grid is in a current imbalance state, it issues a current imbalance alarm based on the alarm level coefficient. The preset trigger probability is set based on actual conditions.

[0044] In summary, the artificial intelligence-based power grid alarm perception system provided by the present invention obtains the electrical physical characteristics and electrical operating status characteristics of the three-phase current by performing feature extraction on the current time series data obtained by three-phase current collection, and performs current imbalance perception through the electrical physical characteristics and electrical operating status characteristics, which can more comprehensively describe the complex working conditions of the power grid; on the other hand, it matches the target current imbalance perception model that best suits the current electrical physical characteristics in the preset model library, and then performs current imbalance perception on the current electrical operating status characteristics through the target current imbalance perception model to obtain accurate current imbalance perception results, thereby accurately performing current imbalance alarm perception on the power grid.

[0045] Example 2, reference Figure 2 , which is a second embodiment of the present invention, provides a power grid alarm perception method based on artificial intelligence, including:

[0046] S1: The three-phase current in the power grid is collected based on distributed sensors to obtain current time series data.

[0047] In an embodiment of the present invention, distributed sensors are deployed at different locations in the power grid. Therefore, the present invention calls on the distributed sensors to sample the three-phase current in the power grid at a specific sampling frequency, and converts the collected analog signals into digital signals to obtain discrete current time series data I(t).

[0048] S2: Perform feature extraction on the current time series data to obtain multi-dimensional feature extraction.

[0049] In an embodiment of the present invention, multi-dimensional feature extraction includes electrical physical features and electrical operating state features.

[0050] Furthermore, the electrical physical characteristics include three-phase current spectrum characteristics, three-phase current transient characteristics and three-phase current high-frequency component characteristics.

[0051] Furthermore, the electrical operating status characteristics include three-phase current amplitude, three-phase current phase and three-phase current harmonic content. Therefore, it is necessary to extract the current spectrum characteristics, current transient characteristics, current high-frequency component characteristics, current amplitude, current phase and current harmonic content from the current time series data.

[0052] In the embodiment of the present invention, for the extraction of the three-phase current amplitude, since the current time series data I(t) includes the three-phase current signal I a (t),I b (t),I c (t), so the three-phase current signal is I a (t),I b (t),I c(t) Discretize within a period T and divide the period T into N equally spaced time points t i ,i=1,2,...,N, so for the three-phase current signal I a (t) the three-phase current amplitude A a The specific extraction formula is as follows:

[0053]

[0054] Among them, α(t) is the coefficient adjusted according to the actual grid environment. Similarly, the three-phase current signal I can be extracted. b (t) the three-phase current amplitude A b And the three-phase current signal I c (t) the three-phase current amplitude A c .

[0055] In the embodiment of the present invention, for extracting the phase of the three-phase current, the three-phase current signal I a (t) and three-phase current signal I b (t) The phase difference between The specific formula is as follows:

[0056]

[0057] Similarly, the three-phase current signal I can be extracted a (t) and three-phase current signal I c (t) The phase difference between Three-phase current signal I b (t) and three-phase current signal I c (t) The phase difference between

[0058] In the embodiment of the present invention, for the three-phase current harmonic content, the current time series data I(t) is decomposed into a plurality of sub-signals I with different frequencies. f (t), the frequency f increases in integer multiples starting from the fundamental frequency, so the harmonic content H of the nth harmonic is n The specific formula is as follows:

[0059]

[0060] Where n is the harmonic order; I nf (t) is the sub-signal corresponding to the nth harmonic frequency decomposed from the current time series data I(t).

[0061] In the embodiment of the present invention, the spectrum characteristics of the three-phase current are analyzed by the spectrum characteristic matrix S(t). The spectrum characteristic matrix S(t) fully describes the distribution and mutual relationship of the three-phase current signal at different frequency components. Each element S in the matrix ij They all carry specific information about the spectral characteristics of the current signal. By analyzing the spectral characteristic matrix S(t) (such as calculating eigenvalues, eigenvectors, etc.), the spectral structure of the current signal can be obtained, such as the distribution and mutual relationship of the current signal at different frequency components. Among them, for the element S in the i-th row and j-th column of the spectral characteristic matrix S(t), ij , the specific formula is as follows:

[0062]

[0063] In the embodiment of the present invention, for the three-phase current transient characteristics, since the transient occurrence time is t0, the current signal during the transient process is I trans (t), therefore, the extracted transient characteristic index T m The specific formula is as follows:

[0064]

[0065] where Δt is an estimate of the transient duration.

[0066] In the embodiment of the present invention, for the high-frequency component characteristics of the three-phase current, the high-frequency component characteristics C h The specific extraction formula is as follows:

[0067]

[0068] Among them, C h It is the high frequency component feature.

[0069] S3: Based on the electrical physical characteristics, the preset model library is matched to obtain the target current imbalance perception model; the target current imbalance perception model is trained based on the sample characteristics and the perception result labels.

[0070] In an embodiment of the present invention, a preset model library is pre-constructed, wherein the preset model library includes multiple current imbalance perception models, each current imbalance perception model has different optimal applicable parameters, and each current imbalance perception model is trained based on sample features and perception result labels. The training process of the current imbalance perception model is shown in steps S3.1 to S3.4.

[0071] Therefore, in an embodiment of the present invention, the electrical physical characteristics are matched in the preset model library, and the target current imbalance perception model that is best applicable to the current electrical physical characteristics is obtained in the preset model library, that is, the target current imbalance perception model that is best applicable to the current three-phase current spectrum characteristics, three-phase current transient characteristics and three-phase current high-frequency component characteristics is matched in the preset model library, as shown in steps S3.5 to S3.7.

[0072] In an embodiment of the present invention, the training constraints for training the target current imbalance sensing model include:

[0073] All loss function values ​​are less than or equal to a preset loss threshold, the loss difference between two adjacent loss function values ​​is less than or equal to a preset difference, and the learning rate in the target optimization function is within a preset range. The target optimization function is constructed based on the model parameters of the deep learning network, and the model parameters include weight parameters and bias parameters.

[0074] In an embodiment of the present invention, training a target current imbalance perception model includes the following steps:

[0075] S3.1: Input the sample features into the deep learning network, obtain the perceptual prediction results of each sample feature output by the deep learning network, and determine the loss function value of each sample feature based on the loss function of the deep learning network and the perceptual prediction results and perceptual result labels of each sample feature.

[0076] In an optional embodiment, the present invention inputs sample features into a deep learning network, and obtains a perceptual prediction result of each sample feature output by the deep learning network.

[0077] Furthermore, based on the loss function of the deep learning network and the perception prediction result and perception result label of each sample feature, the loss function value of each sample feature is determined. The specific formula of the loss function is as follows:

[0078]

[0079] in, is the loss function value of the i-th sample feature; y i is the perception result label of the i-th sample feature; is the perceptual prediction result of the i-th sample feature.

[0080] In the embodiment of the present invention, the specific formula of the target optimization function is as follows:

[0081]

[0082] Among them, θ t+1 is the model parameter of the t+1th iteration; θ tis the model parameter of the tth iteration; η t is the learning rate of the tth iteration; m t is the momentum term of the t-th iteration; δ is the preset adjustment parameter; is the gradient of the loss function at the tth iteration; J(θ) is the loss function; is the loss function value of the sample feature; θ is the parameter vector of the model; θ k is the kth parameter vector; λ is the regularization parameter; |θ k | 1.5 is the regularization term.

[0083] S3.2: If there is at least one loss function value greater than a preset loss threshold, or / and, there is at least one loss difference greater than a preset difference, or / and, the learning rate of the deep learning network is outside a preset range, then the model update coefficient is determined based on the perception prediction result and perception result label of each sample feature.

[0084] Furthermore, all loss function values ​​and the learning rate of the deep learning network are constrained and judged according to the training constraints. If it is determined that at least one loss function value is greater than a preset loss threshold, or / and at least one loss difference is greater than a preset difference, or / and the learning rate of the deep learning network is outside a preset range, then a model update coefficient is determined according to the perception prediction result and the perception result label of each sample feature, including the following steps:

[0085] The perception result label y based on the i-th sample feature i and perception prediction results Calculate the sample-level error metric for the i-th sample feature

[0086] Furthermore, according to the sample-level error metric e of the i-th sample feature i , calculate the sample mean error where N s is the total number of sample features.

[0087] Furthermore, according to the sample-level error metric e of the i-th sample feature i And the sample error mean μ e , calculate the sample error standard deviation σ e and sample error entropy H e , the specific formula is as follows:

[0088]

[0089] Among them, e j is the sample-level error measure of the j-th sample feature; p iis the sample-level error metric e of the i-th sample feature i The probability of being in the sum of all error metrics.

[0090] Furthermore, according to the sample error mean μ e , sample error standard deviation σ e and sample error entropy H e , calculate the model update coefficient γ, the specific formula is as follows:

[0091] γ=exp(-(μ e / σ e ))*(1+H e ).

[0092] S3.3: Update the learning rate and function parameters of the target optimization function based on the model update coefficient to obtain the updated deep learning network.

[0093] Furthermore, the learning rate and function parameters of the target optimization function are updated based on the model update coefficient to obtain the updated deep learning network, including:

[0094] The specific formula for learning rate update is as follows:

[0095] η t+1 =η0*γ βt *[1-exp(-(t / T))].

[0096] Among them, η t+1 is the learning rate at the t+1th iteration; η0 is the initial learning rate; β∈[0,1] is the parameter for adjusting the learning rate attenuation; t is the current number of iterations; T is the maximum number of iterations.

[0097] The specific formula for updating function parameters is as follows:

[0098]

[0099] Among them, θ t+1 is the model parameter vector after the t+1th iteration update; θ t is the model parameter vector at the tth iteration; is the gradient vector of the loss function with respect to the model parameter θ at the tth iteration; * is the adjustment parameter; sign(θ t ) is a sign function, when θ t When the element of is greater than 0, it returns 1. t When the element of is less than 0, it returns -1. t When the element of is equal to 0, it returns 0; t | 0.5 is the parameter vector θ tTake the absolute value of each element and then take the square root.

[0100] S3.4: Input the sample features into the updated deep learning network, and obtain the perception prediction results of each sample feature output by the updated deep learning network, until the loss function value obtained based on the loss function and the perception prediction results and perception result labels of each sample feature, and the learning rate of the updated deep learning network meet the training constraints, and the target current imbalance perception model is obtained.

[0101] Furthermore, the sample features are input into the updated deep learning network, and the perception prediction results of each sample feature output by the updated deep learning network are obtained, until the loss function value obtained based on the loss function and the perception prediction results and perception result labels of each sample feature, and the learning rate of the updated deep learning network meet the training constraints, that is, all loss function values ​​are less than or equal to the preset loss threshold, the loss difference between two adjacent loss function values ​​is less than or equal to the preset difference, and the learning rate in the target optimization function is within the preset range, thereby obtaining the target current imbalance perception model.

[0102] It should be noted that the embodiment of the present invention trains a target current imbalance perception model, and thus uses the target current imbalance perception model to perform current imbalance perception on the current electrical operating status characteristics, thereby obtaining accurate current imbalance perception results, thereby accurately perceiving current imbalance alarms on the power grid.

[0103] Furthermore, the target current imbalance perception model that best suits the current three-phase current spectrum characteristics, three-phase current transient characteristics, and three-phase current high-frequency component characteristics is matched in the preset model library, including the following steps:

[0104] S3.5: Based on the three-phase current spectrum characteristics, the three-phase current transient characteristics and the three-phase current high-frequency component characteristics, determine the comprehensive matching value of each current imbalance perception model in the preset model library to the electrical and physical characteristics of the power grid.

[0105] In an embodiment of the present invention, determining a comprehensive matching value of each current imbalance perception model in a preset model library to the electrical and physical characteristics of a power grid includes the following steps:

[0106] Obtain the optimal spectrum characteristics, optimal transient characteristics and optimally adapted high-frequency characteristics of each current imbalance perception model for three-phase current in the preset model library.

[0107] According to the three-phase current spectrum characteristics and the optimal spectrum characteristics of each current imbalance perception model for the three-phase current, the similarity of the spectrum characteristics of each current imbalance perception model is determined, including: the vector of the extracted three-phase current spectrum characteristics is S = [s1, s2, ..., s p], where p is the vector dimension, and the vector of the optimal spectrum characteristics of the kth current unbalance sensing model for the three-phase current is Y k =[y k1 ,y k2 ,...,y kp ].

[0108] Furthermore, the vector S of the three-phase current spectrum characteristics and the vector Y of the optimal spectrum characteristics of the three-phase current of the kth current unbalance sensing model are calculated. k The local correlation coefficient between them is as follows:

[0109]

[0110] Among them, C ij For vector S and vector Y k The local correlation coefficient in the local window with a size of w*w starting at (M,Q); w is the local window size; m and q are the variables traversed and summed in the local window with (M,Q) as the starting point; s M is the value of vector S at point M; is the average value of the elements of vector S in the local window; y kM is the vector Y k The value at point M; is the vector Y k The average value of the elements in the local window.

[0111] Furthermore, according to the local correlation coefficient C ij Calculate the similarity of the spectrum characteristics of each current imbalance perception model. The specific formula is as follows:

[0112]

[0113] Among them, Sim k is the similarity of the spectrum characteristics of the kth current imbalance perception model; p is the vector dimension.

[0114] According to the three-phase current transient characteristics and the optimal transient characteristics of each current imbalance perception model for the three-phase current, the current transient matching value of each current imbalance perception model is determined, including: the vector of the extracted three-phase current transient characteristics is A=[a1,a2,...,a z ], where z is the vector dimension, and the vector of the optimal transient characteristics of the kth current unbalance sensing model for the three-phase current is B k =[b k1 ,b k2 ,...,b kz ].

[0115] The vector A of the three-phase current transient characteristics and the vector B of the optimal transient characteristics of the three-phase current of the kth current unbalance perception model are combined. k Perform wavelet transform to obtain coefficients W at different scales A and

[0116] Furthermore, the coefficient W at different scales is calculated A and The transient feature W at the i-th scale and j-th position Aij and The energy difference E ij , the specific formula is as follows:

[0117]

[0118] According to the energy difference E of transient characteristics ij Calculate the current transient matching value of each current imbalance sensing model. The specific formula is as follows:

[0119]

[0120] Among them, M k is the current transient matching value of the kth current unbalance sensing model; h(i,j) is the weight coefficient; v is the scale of the wavelet transform; and z is the vector dimension.

[0121] Furthermore, according to the high-frequency component characteristics of the three-phase current and the best-adapted high-frequency characteristics of each current imbalance perception model to the three-phase current, the high-frequency characteristic adaptation value of each current imbalance perception model is determined, including: the vector of the extracted high-frequency component characteristics of the three-phase current is C = [c1, c2, ..., c L ], where L is the vector dimension, and the vector of the best adaptation high-frequency characteristics of the kth current unbalance sensing model to the three-phase current is D k =[d k1 ,d k2 ,...,d kL ].

[0122] The first high-order cumulative quantity is calculated based on the vector C of the high-frequency component characteristics of the three-phase current, and the vector D of the best-adapted high-frequency characteristics of the three-phase current according to the k-th current imbalance perception model is calculated based on the vector C of the high-frequency component characteristics of the three-phase current. k Calculate the second-highest-order cumulant. The specific formula is as follows:

[0123]

[0124] Among them, T ij is the first high-order cumulant; T kij is the second-highest-order cumulant; E[·] is the mathematical expectation; c iis the i-th vector in the vector of the extracted three-phase current high-frequency component characteristics; c j is the jth vector in the vector of the extracted three-phase current high-frequency component characteristics; c i+1 is the i+1th vector in the vector of the extracted three-phase current high-frequency component characteristics; d ki is the i-th vector in the vector of the best adaptation of the high-frequency characteristics of the three-phase current by the k-th current unbalance sensing model; d kj is the jth vector in the vector of the best adaptation of the high-frequency characteristics of the three-phase current by the kth current unbalance sensing model; d k(i+1) It is the i+1th vector among the vectors that best adapt the high-frequency characteristics of the three-phase current to the kth current unbalance sensing model.

[0125] According to the first high-order cumulant T ij and the second high-order cumulant T kij , calculate the high-frequency feature adaptation value of each current imbalance perception model. The specific formula is as follows:

[0126]

[0127] Among them, F k is the high-frequency feature adaptation value of the k-th current imbalance perception model; L is the vector dimension.

[0128] Further, according to the spectrum characteristic similarity, current transient matching value and high frequency characteristic adaptation value of each current imbalance perception model, the comprehensive matching value of each current imbalance perception model is determined, including: according to the spectrum characteristic similarity Sim of the kth current imbalance perception model k , current transient matching value M k and high-frequency feature adaptation value F k , construct the three-dimensional space vector V of the k-th current unbalance perception model k =[Sim k ,M k ,F k ].

[0129] Calculate the three-dimensional space vector V of the kth current unbalance sensing model k and the preset ideal matching vector V ideal = angle θ of [1,1,1] k , the specific formula is as follows:

[0130]

[0131] According to the angle θ of the kth current unbalance sensing model k , calculate the comprehensive matching value of the kth current imbalance sensing model. The specific formula is as follows:

[0132]

[0133] Among them, Z k is the comprehensive matching value of the kth current unbalance sensing model.

[0134] S3.6: Traverse the comprehensive matching value of each current imbalance sensing model to obtain the maximum comprehensive matching value.

[0135] S3.7: Determine the current imbalance sensing model corresponding to the maximum comprehensive matching value as the target current imbalance sensing model.

[0136] It should be noted that by traversing the comprehensive matching value of each current imbalance perception model, the maximum comprehensive matching value is obtained, and the current imbalance perception model corresponding to the maximum comprehensive matching value is determined as the target current imbalance perception model, wherein the target current imbalance perception model that is best suitable for the current three-phase current spectrum characteristics, three-phase current transient characteristics and three-phase current high-frequency component characteristics is matched in the preset model library, and then the current imbalance perception of the current electrical operation status characteristics is performed through the target current imbalance perception model to obtain accurate current imbalance perception results, thereby accurately perceiving the current imbalance alarm of the power grid; on the other hand, the complex working conditions of the power grid are more comprehensively described through the three-phase current spectrum characteristics, three-phase current transient characteristics and three-phase current high-frequency component characteristics.

[0137] S4: Inputting the electrical operation state characteristics into the target current imbalance perception model to obtain a current imbalance perception result output by the target current imbalance perception model.

[0138] In an embodiment of the present invention, the current electrical operating status characteristics are input into the target current imbalance perception model. The target current imbalance perception model performs current imbalance perception prediction based on the current three-phase current amplitude, three-phase current phase and three-phase current harmonic content, and outputs the current imbalance perception result, as shown in steps S4.1 to S4.3.

[0139] In an embodiment of the present invention, outputting a current imbalance sensing result includes the following steps:

[0140] S4.1: Input the three-phase current amplitude, three-phase current phase and three-phase current harmonic content into the target current imbalance perception model. For the data processing layer, generate an amplitude matrix according to the three-phase current amplitude and three-phase current phase, and generate a harmonic matrix according to the three-phase current harmonic content.

[0141] In an embodiment of the present invention, the target current imbalance perception model includes a data processing layer and a model prediction layer. Therefore, the three-phase current amplitude, three-phase current phase and three-phase current harmonic content are input into the target current imbalance perception model, and the data processing layer generates an amplitude matrix based on the three-phase current amplitude and three-phase current phase. In one embodiment, the three-phase current amplitude is [A a ,A b ,A c ], the three-phase current phase is Amplitude matrix M A The specific formula is as follows:

[0142]

[0143] Furthermore, the harmonic content of the three-phase current at different harmonic orders is the harmonic content H a,n , harmonic content H a,n and harmonic content H c,n (n is the harmonic number), the data processing layer is based on the harmonic content H n,a , harmonic content H n,b and harmonic content H n,c Construct harmonic matrix M H , where the harmonic matrix M H It is a 3*Num dimensional matrix, Num is the total number of harmonics, so for the harmonic matrix M H Each element M in Hij (i=a,b,c;j=1,2,...,Num) can be expressed as:

[0144]

[0145] in, is the sum of the three-phase harmonic content under the jth harmonic; H ij is the i-th phase harmonic content under the j-th harmonic.

[0146] S4.2: For the model prediction layer, the amplitude matrix and the harmonic matrix are fused to obtain a fused tensor, and the fused tensor is transformed to obtain a transformed result.

[0147] Furthermore, in the embodiment of the present invention, the amplitude matrix and the harmonic matrix are fused based on the singular value decomposition (SVD) fusion algorithm. Therefore, the model prediction layer calculates the amplitude matrix M A Perform singular value decomposition and the harmonic matrix M H Perform singular value decomposition, that is Among them, U A It is an orthogonal matrix, and the column vector is called the left singular vector, which is used to describe the amplitude matrix M A The orthogonal basis of the row space of Σ AIt is a diagonal matrix, and the elements on the diagonal (singular values) are arranged from large to small; is an orthogonal matrix, and the column vector is called the right singular vector, which is used to describe the amplitude matrix M A The orthogonal basis of the column space. Similarly, we can understand the parameter U H , parameter Σ H and parameters The meaning of is not repeated here.

[0148] Furthermore, the model prediction layer is based on the amplitude matrix M A Σ after singular value decomposition A and harmonic matrix M H Σ after singular value decomposition H Construct the fusion coefficient matrix M V , where for the fusion coefficient matrix M C Each element V in ij It can be expressed as:

[0149]

[0150] Among them, σ Aij and σ Hij ∑ A and M H The element in row i and column j of Akl and σ Hkl Σ A and M H The element at row k and column l of .

[0151] Furthermore, the model prediction layer calculates the fused tensor T fused , the specific formula is as follows:

[0152]

[0153] Among them, V 11 is the fusion coefficient matrix M C The first row and first column element in V 12 is the fusion coefficient matrix M C The first row and second column element in V mn is the fusion coefficient matrix M C The element in row m and column n; V nm is the fusion coefficient matrix M C The element at row n and column m in .

[0154] Furthermore, the model prediction layer is fused and the tensor is transformed. The embodiment of the present invention is based on the transformation function of the convolutional neural network (CNN) F Transform the fused tensor, where the transformation function trans FThe convolution kernel size is K i ,i=1,2,...,Sum, Sum is the number of convolution layers; K i To dynamically change the function according to the input features, in is the input of the i-th convolution layer, is the sum of the elements.

[0155] Therefore, after the Sum layer convolution operation and activation function processing, the transformed result Return = transform F (T fused ), where the activation function can be expressed as

[0156] S4.3: Perform current imbalance sensing based on the transformed results, the fused tensor, and the three-phase current amplitudes, and output the current imbalance sensing results.

[0157] In an embodiment of the present invention, the model prediction layer performs current imbalance perception based on the transformed results, the fused tensor and the three-phase current amplitude, and outputs the current imbalance perception result, as specifically shown in steps S4.3.1 to S4.3.3.

[0158] It should be noted that the present invention uses a target current imbalance perception model to analyze the current three-phase current amplitude, phase, and harmonic content to perform current imbalance sensing, obtaining accurate current imbalance sensing results and thus accurately detecting current imbalance alarms in the power grid. Furthermore, the three-phase current amplitude, phase, and harmonic content provide a more comprehensive description of the complex operating conditions of the power grid.

[0159] Furthermore, the model prediction layer performs current imbalance perception based on the transformed results, the fused tensor, and the three-phase current amplitudes, and outputs the current imbalance perception results, including the following steps:

[0160] S4.3.1: Determine the current unbalance probability based on the transformed results.

[0161] In the embodiment of the present invention, the model prediction layer calculates a probability value, i.e., a current imbalance probability, based on the transformed result. The current imbalance probability represents the possibility that the power grid is in a current imbalance state. The specific formula for the current imbalance probability is as follows:

[0162]

[0163] Where P(R) is the current unbalance probability; Return is the result after transformation.

[0164] S4.3.2: Determine the alarm level coefficient for current imbalance based on the transformed results, the fused tensor, and the three-phase current amplitudes.

[0165] Furthermore, the model prediction layer calculates the transformed result Return and the fused tensor T fused Mutual information MI(Return,T fused ), the specific formula is as follows:

[0166]

[0167] Where, p(Return=r,T fused =t) is the transformed result Return and the fused tensor T fused The joint probability distribution of p(Return=r) is the marginal probability distribution of the transformed result Return; p(T fused =t) is the fused tensor T fused The marginal probability distribution of ; the above joint probability distribution and marginal probability distribution can be calculated based on the kernel density estimation method, which will not be repeated here.

[0168] Furthermore, according to the model prediction layer, the mutual information MI (Return, T fused ) Calculate the transformed result Return and the fused tensor T fused The similarity between RT , the specific formula is as follows:

[0169]

[0170] Furthermore, the model prediction layer is based on the three-phase current amplitude [A a ,A b ,A c ]Calculate the amplitude difference measure D A , amplitude difference measure D A The specific formula is as follows:

[0171]

[0172] Furthermore, the model prediction layer is based on the similarity S RT and amplitude difference measure D A Calculate the alarm level coefficient for current imbalance. The specific formula is as follows:

[0173] Alarmlevel=S RT / (1+D A ).

[0174] S4.3.3: Determine the current imbalance sensing result based on the current imbalance probability and the alarm level coefficient.

[0175] Furthermore, the model prediction layer outputs the current imbalance probability and the alarm level coefficient as the current imbalance perception result.

[0176] In an optional embodiment, the current imbalance sensing result 1 may be {current imbalance probability: 30%; alarm level coefficient: 2}, the current imbalance sensing result 2 may be {current imbalance probability: 70%; alarm level coefficient: 7}, and so on.

[0177] It should be noted that the present invention performs current imbalance perception based on the transformed results, the fused tensor and the three-phase current amplitude, and can accurately obtain the current imbalance probability and the alarm level coefficient, thereby realizing accurate current imbalance alarm perception of the power grid, enabling maintenance personnel to take corresponding measures according to the alarm level coefficient, and ensuring the stable operation of the power grid.

[0178] S5: Based on the current imbalance sensing result, a current imbalance alarm is issued to the power grid.

[0179] In an embodiment of the present invention, the current imbalance sensing result includes a current imbalance probability and a corresponding alarm level coefficient. The current imbalance probability represents the likelihood that the power grid is in a current imbalance state. Therefore, the present invention analyzes the current imbalance sensing result and compares the resulting current imbalance probability with a preset trigger probability to obtain a comparison result. The preset trigger probability is set based on actual conditions; a larger alarm level coefficient indicates a higher alarm level.

[0180] If it is determined that the current imbalance probability is greater than the preset trigger probability, that is, when it is determined that the power grid is in a current imbalance state, the present invention issues a current imbalance alarm according to the alarm level coefficient.

[0181] It should be noted that by extracting features from the current time series data obtained from three-phase current collection, the electrical physical characteristics and electrical operating status characteristics of the three-phase current can be obtained. Current imbalance perception is performed through the electrical physical characteristics and electrical operating status characteristics, which can more comprehensively describe the complex working conditions of the power grid. On the other hand, the target current imbalance perception model that best suits the current electrical physical characteristics is matched in the preset model library, and then the current imbalance perception is performed on the current electrical operating status characteristics through the target current imbalance perception model to obtain accurate current imbalance perception results, thereby accurately perceiving the current imbalance alarm of the power grid.

[0182] Embodiment 3 is the third embodiment of the present invention, which differs from the first two embodiments in that:

[0183] like Figure 3 and Figure 4As shown, if the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0184] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0185] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0186] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0187] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. The power grid alarm perception system based on artificial intelligence is characterized by: include, A power grid alarm perception center, a data acquisition module, a feature extraction module, a model matching module, a model prediction module, and a current imbalance alarm perception module; the power grid alarm perception center is respectively connected to the data acquisition module, the feature extraction module, the model matching module, the model prediction module, and the current imbalance alarm perception module to manage each module; A data acquisition module is used to collect three-phase currents in the power grid based on distributed sensors to obtain current time series data; A feature extraction module is used to extract features from the current time series data to obtain multi-dimensional feature extraction; the multi-dimensional feature extraction includes electrical physical features and electrical operating state features; A model matching module is configured to match a preset model library based on the electrical physical characteristics to obtain a target current imbalance perception model; the target current imbalance perception model is trained based on sample characteristics and perception result labels; a model prediction module, configured to input the electrical operating state characteristics into the target current imbalance perception model to obtain a current imbalance perception result output by the target current imbalance perception model; The current imbalance alarm sensing module is used to issue a current imbalance alarm to the power grid based on the current imbalance sensing result.

2. An artificial intelligence-based power grid alarm perception method, using the artificial intelligence-based power grid alarm perception system according to claim 1, characterized in that: include: The three-phase current in the power grid is collected based on distributed sensors to obtain current time series data; Performing feature extraction on the current time series data to obtain multi-dimensional feature extraction; the multi-dimensional feature extraction includes electrical physical features and electrical operating state features; Based on the electrical physical characteristics, a target current imbalance perception model is obtained by matching in a preset model library; the target current imbalance perception model is obtained by training based on sample characteristics and perception result labels; Inputting the electrical operating state characteristics into the target current imbalance sensing model to obtain a current imbalance sensing result output by the target current imbalance sensing model; A current imbalance alarm is issued to the power grid based on the current imbalance sensing result.

3. The artificial intelligence-based power grid alarm perception method according to claim 2, characterized in that: The electrical physical characteristics include three-phase current spectrum characteristics, three-phase current transient characteristics and three-phase current high-frequency component characteristics; Based on the electrical and physical characteristics, a target current imbalance sensing model is obtained by matching in a preset model library, including: Determining a comprehensive matching value of each current imbalance perception model in the preset model library to the electrical and physical characteristics of the power grid based on the three-phase current spectrum characteristics, the three-phase current transient characteristics, and the three-phase current high-frequency component characteristics; Traverse the comprehensive matching values ​​of each current imbalance sensing model to obtain the maximum comprehensive matching value; The current imbalance sensing model corresponding to the maximum comprehensive matching value is determined as the target current imbalance sensing model.

4. The artificial intelligence-based power grid alarm perception method according to claim 3, characterized in that: Determining a comprehensive matching value of each current imbalance perception model in the preset model library to the electrical and physical characteristics of the power grid based on the three-phase current spectrum characteristics, the three-phase current transient characteristics, and the three-phase current high-frequency component characteristics includes: Determining a similarity between the spectrum characteristics of each current imbalance perception model and the optimal spectrum characteristics of the three-phase current of each current imbalance perception model; Determining a current transient matching value of each current imbalance sensing model based on the three-phase current transient characteristics and the optimal transient characteristics of each current imbalance sensing model for the three-phase current; Determining a high-frequency feature adaptation value of each current imbalance perception model based on the high-frequency component characteristics of the three-phase current and the best-adapted high-frequency characteristics of each current imbalance perception model to the three-phase current; Based on the spectrum characteristic similarity, current transient matching value and high-frequency characteristic adaptation value of each current imbalance sensing model, a comprehensive matching value of each current imbalance sensing model is determined.

5. The artificial intelligence-based power grid alarm perception method according to claim 4, characterized in that: The electrical operation status characteristics include three-phase current amplitude, three-phase current phase and three-phase current harmonic content; The target current imbalance perception model includes a data processing layer and a model prediction layer; Inputting the electrical operating state characteristics into the target current imbalance sensing model to obtain a current imbalance sensing result output by the target current imbalance sensing model includes: Inputting the three-phase current amplitudes, the three-phase current phases, and the three-phase current harmonic content into the target current imbalance perception model, and generating an amplitude matrix according to the three-phase current amplitudes and the three-phase current phases, and generating a harmonic matrix according to the three-phase current harmonic content at the data processing layer; For the model prediction layer, the amplitude matrix and the harmonic matrix are fused to obtain a fused tensor, and the fused tensor is transformed to obtain a transformed result; Current imbalance sensing is performed based on the transformed result, the fused tensor and the three-phase current amplitudes, and the current imbalance sensing result is output.

6. The artificial intelligence-based power grid alarm perception method according to claim 5, characterized in that: Performing current imbalance sensing based on the transformed result, the fused tensor, and the three-phase current amplitudes, and outputting the current imbalance sensing result, including: Determining a current imbalance probability based on the transformed result; the current imbalance probability represents the possibility that the power grid is in a current imbalance state; Determining an alarm level coefficient of current imbalance based on the transformed result, the fused tensor, and the three-phase current amplitudes; A current imbalance sensing result is determined based on the current imbalance probability and the alarm level coefficient.

7. The artificial intelligence-based power grid alarm perception method according to claim 6, characterized in that: The training constraints are: all loss function values ​​are less than or equal to a preset loss threshold, the difference between two adjacent loss function values ​​is less than or equal to a preset difference, and the learning rate in the target optimization function is within a preset range. The target optimization function is constructed based on the model parameters of the deep learning network, including weight parameters and bias parameters. Training the target current imbalance perception model includes the following steps: Input the sample features into the deep learning network, obtain the perception prediction results of each sample feature output by the deep learning network, and determine the loss function value of each sample feature based on the loss function of the deep learning network and the perception prediction results and perception result labels of each sample feature; If at least one loss function value is greater than a preset loss threshold, or / and, at least one loss difference value is greater than a preset difference value, or / and, the learning rate of the deep learning network is outside a preset range, then a model update coefficient is determined based on the perception prediction result and the perception result label of each sample feature; Updating the learning rate and function parameters of the target optimization function based on the model update coefficient to obtain an updated deep learning network; The sample features are input into the updated deep learning network, and the perception prediction results of each sample feature output by the updated deep learning network are obtained, until the loss function value obtained based on the loss function and the perception prediction results and perception result labels of each sample feature, and the learning rate of the updated deep learning network meet the training constraints, and the target current imbalance perception model is obtained.

8. The artificial intelligence-based power grid alarm perception method according to claim 7, characterized in that: The specific formula of the loss function is as follows: in, is the loss function value of the i-th sample feature; y i is the perception result label of the i-th sample feature; is the perceptual prediction result of the i-th sample feature; The specific formula of the objective optimization function is as follows: Among them, θ t+1 is the model parameter of the t+1th iteration; θ t is the model parameter of the tth iteration; η t is the learning rate of the tth iteration; m t is the momentum term of the t-th iteration; δ is the preset adjustment parameter; is the gradient of the loss function at the tth iteration; J(θ) is the loss function; is the loss function value of the sample feature; θ is the parameter vector of the model; θ k is the kth parameter vector; λ is the regularization parameter; |θ k | 1.5 is the regularization term.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the artificial intelligence-based power grid alarm perception system according to claim 1 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence-based power grid alarm perception system according to claim 1 are implemented.

Images