Safe power utilization system based on humid environment short circuit and harmonic suppression

Abstract

The invention belongs to the technical field of power utilization management, and discloses a safe power utilization system based on humid environment short circuit and harmonic suppression. Comprising the following steps: collecting environmental data and power data of a power system, and preprocessing the environmental data and the power data to obtain a plurality of segments of processed environmental data and processed power data; calculating a plurality of short circuit risk indexes and harmonic content indexes according to the processed environment data and the processed power data; according to the short circuit risk index and the harmonic content index, constructing a multi-objective optimization model, and based on the constructed multi-objective optimization model, solving a plurality of corresponding optimal setting electric power parameters; the solved optimal set electric power parameters are applied to the electric power system, the electricity utilization safety and reliability can be remarkably improved, the energy consumption can be reduced, and the economic loss is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of power management, and more specifically, to a safe power system based on short circuit and harmonic control in a humid environment. Background Art

[0002] The patent with application publication number CN115133532A discloses a method, device, equipment, storage medium and storage medium for controlling an electric power system; including: obtaining object data of an equipment entity object in the electric power system, and obtaining business data in an electric power business system corresponding to the electric power system; constructing a digital twin electric power system model based on the object data and the business data; obtaining system optimization and reconstruction requirements, and generating optimization parameters based on the system optimization and reconstruction requirements, and constructing an optimization simulation environment for the digital twin electric power system model based on the optimization parameters; simulating the digital twin electric power system model based on the optimization simulation environment, and obtaining simulation results; and feeding back the optimal parameters in the simulation results to the electric power system and the electric power business system to control and process the electric power system and the electric power business system, so as to achieve control over the electric power system.

[0003] However, the existing power consumption system still has significant shortcomings in fully ensuring safety and power quality, especially when dealing with short-circuit risks and harmonic interference in humid environments. For example, in complex industrial environments, equipment aging and environmental factors often lead to short-circuit accidents and harmonic pollution, affecting production efficiency and safety. At the same time, it is difficult to maintain stable operation in a changing environment. For example, the high humidity and high salt spray environment of coastal chemical plants increase safety hazards. Traditional data processing methods are relatively rough and lack the ability to accurately capture data characteristics and changing trends, resulting in the inability to timely detect potential faults in places such as large data centers. In addition, due to the lack of segmented differentiated analysis capabilities, existing systems find it difficult to achieve accurate load forecasting and scheduling in smart grids, nor can they provide personalized power consumption strategies for large complexes. Existing technologies generally lack multi-objective optimization and adaptive capabilities, making it difficult to achieve a balance under multiple factors. These problems have led to energy waste, increased operating costs, and limited the overall performance and safety and reliability of the power system.

[0004] In view of this, the present invention proposes a safe power consumption system based on short circuit and harmonic control in humid environment to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solutions: a safe power system based on short circuit and harmonic control in a humid environment, comprising: a data acquisition and processing module, an exponential fitting module and a strategy optimization module; each module is connected by wire and / or wireless means;

[0006] The data acquisition and processing module includes an acquisition unit and a processing unit; the acquisition unit is used to acquire environmental data and power data of the power system, and the processing unit is used to pre-process the environmental data and power data to obtain several segments of processed environmental data and processed power data;

[0007] The index fitting module is used to calculate several short circuit risk indexes and harmonic content indexes based on the processed environmental data and the processed power data;

[0008] The strategy optimization module includes a model construction unit and an optimization solution unit; the model construction unit is used to construct a multi-objective optimization model according to the short-circuit risk index and the harmonic content index; the optimization solution unit solves several corresponding optimal setting power parameters based on the constructed multi-objective optimization model; and the solved optimal setting power parameters are applied to the power system.

[0009] Furthermore, the environmental data includes environmental temperature, environmental humidity and environmental dust concentration; for a seaside environment, the environmental data also includes environmental salt spray concentration; the power data includes voltage, current, insulation resistance, leakage current and load type.

[0010] Furthermore, the method of preprocessing the environmental data and the power data includes:

[0011] The environmental data and the power data are sorted into several time series in chronological order, and each time series contains all the environmental data and power data at a specific time point; several time series are combined side by side by columns to construct a two-dimensional data matrix, and the rows of the two-dimensional data matrix represent time points, and the columns represent different environmental data and power data; a column is added to the two-dimensional data matrix to represent the timestamp corresponding to each row, and several columns are added to the two-dimensional data matrix to represent the spatial coordinates corresponding to each row of data, so as to obtain a multi-dimensional data matrix; the multi-dimensional data matrix is ​​standardized to obtain a standard multi-dimensional data matrix; the standard multi-dimensional data matrix is ​​preliminarily analyzed to obtain several multi-dimensional data groups; the time distribution characteristics and spatial distribution characteristics of the data objects within each multi-dimensional data group are obtained; according to the time distribution characteristics and spatial distribution characteristics, each multi-dimensional data group is subdivided into several segments; according to the determined segment boundaries, the data objects of each segment are extracted from the standard multi-dimensional data matrix; several segments of processed environmental data and several groups of processed power data are obtained.

[0012] Furthermore, the method of performing preliminary analysis on the standard multi-dimensional data matrix includes:

[0013] Expand the standard multi-dimensional data matrix into several data vectors by row, regard each data vector as a waveform signal, and construct several waveform data; convolve the waveform data with Coiflet wavelet basis functions at different scales to obtain wavelet decomposition coefficients of corresponding scales; quantize the wavelet decomposition coefficients to obtain quantized wavelet decomposition coefficients, and pack all the quantized wavelet decomposition coefficients in a packing order to obtain conversion domain coefficients;

[0014] Extract eigenvalues ​​from the conversion domain coefficients, including energy, entropy and peak value; construct the extracted eigenvalues ​​into new eigenvectors in the feature space, and the eigenvectors corresponding to all data objects form a feature point cloud in the feature space;

[0015] In the feature space, density estimation is performed on the neighborhood of each feature point to obtain the density estimation value corresponding to each feature point; the density estimation value corresponding to each feature point is normalized to obtain the corresponding density value; a density threshold is preset, and the feature point with a density value greater than the density threshold is used as the center point of the initial multi-dimensional data cluster; for feature points that are not center points, they are assigned to the multi-dimensional data cluster where the nearest center point is located to obtain preliminary multi-dimensional data clusters; all preliminary multi-dimensional data clusters are post-processed to obtain several final multi-dimensional data clusters.

[0016] Furthermore, the method of quantizing the wavelet decomposition coefficients includes:

[0017] Define the sparse representation of wavelet decomposition coefficients, that is, wavelet decomposition coefficients in, is the wavelet basis matrix, α is the sparse coefficient vector to be determined; the sparse representation is reconstructed to obtain the measurement process function, and the formula of the measurement process function is: Among them, θ is the measurement matrix and y is the observation value;

[0018] Define the optimization model including objective function FH and constraint function;

[0019] FH=min∑w_i×|α(i)|0; the formula of constraint function is: Where w_i is the weight of the i-th sparse coefficient vector, α(i) is the i-th sparse coefficient vector; |α(i)|0 is the L0 norm of α(i);

[0020] Solve the optimization model to obtain the optimal sparse coefficients; retain the non-zero elements in the optimal sparse coefficients, quantize the remaining elements to 0, and obtain the quantized sparse coefficient vector as the quantized wavelet decomposition coefficients;

[0021] The method of solving the optimization model includes:

[0022] Initialize a sparse solution α_0, which takes a vector of all zeros or a random vector; calculate the initial residual

[0023] For the kth iteration; calculate the gradient vector in, is the transpose of the wavelet basis matrix, θ T is the transpose of the measurement matrix; r_(k-1) is the residual of the k-1th iteration;

[0024] The gradient vector is thresholded to obtain the intermediate solution α'_k. The formula for thresholding is: α'_k=α_(k-1)+g_k; where α_(k-1) is the sparse solution obtained at the k-1th iteration;

[0025] Keep the k elements with the largest absolute values ​​in α'_k, set the rest to 0, and get a new sparse solution α_k; calculate a new residual based on the new sparse solution α_k The iterative process continues until the preset number of iterations is reached, and the last obtained sparse solution is output as the optimal sparse coefficient.

[0026] Furthermore, the method of post-processing all preliminary multi-dimensional data groups includes:

[0027] For each preliminary multi-dimensional data cluster, a corresponding weighted undirected graph G = (V, E) is constructed; V is a set of nodes, and E is a set of edges; each feature point in the preliminary multi-dimensional data cluster is used as a node of the weighted undirected graph G; the cosine similarity of any two feature points is calculated, and a similarity threshold is preset. If the cosine similarity is greater than the similarity threshold, an edge is connected between the nodes corresponding to the two feature points; the weight of the edge is the cosine similarity between the two feature points;

[0028] Calculate the degree matrix D and adjacency matrix A of the weighted undirected graph G corresponding to the preliminary multi-dimensional data group, and calculate the Laplacian matrix L = DA;

[0029] Calculate the vectors corresponding to the first K smallest non-zero eigenvalues ​​of the Laplace matrix L, map each feature point in the preliminary multidimensional data cluster to the K-dimensional space formed by these K vectors, and perform K-Means clustering on the feature points in the preliminary multidimensional data cluster in the K-dimensional space, that is, split the preliminary multidimensional data cluster into several subclusters, and all the obtained subclusters are multidimensional data clusters.

[0030] Furthermore, the acquisition method of the time distribution characteristics and the space distribution characteristics includes:

[0031] For each data object in a multi-dimensional data group, extract its timestamp information and construct a data time series; for each data object in a multi-dimensional data group, extract its spatial coordinate information; and calculate the spatial distance between data objects to construct a distance matrix;

[0032] Perform discrete Fourier transform on the data time series to obtain the amplitude spectrum in the frequency domain, and calculate the power spectrum density based on the amplitude spectrum. Where N1 is the length of the data time series, X(e) is the complex coefficient of the e-th frequency component obtained by discrete Fourier transform of the data time series, that is, the amplitude spectrum;

[0033] The calculated power spectrum density constitutes a power spectrum density graph. The peak value is found in the power spectrum density graph, and the frequency corresponding to the peak value is taken as the periodic component, which is the time distribution feature.

[0034] Take any two data objects in the distance matrix as data pairs, count the distances between all data pairs, and use them as weight parameters between data pairs;

[0035] Initialize each data object in the distance matrix as a tree containing only itself to form a forest F. Start from the data pair with the smallest weight parameter and traverse all data pairs. If the current data pair belongs to different trees in F, add this data pair to F and merge the two trees into one tree. If the current data pair belongs to the same tree in F, skip the data pair. Repeat until there is only one tree left in F. At this time, this tree is recorded as a greedy tree.

[0036] Calculate the topological index of the greedy tree as the spatial distribution feature. The topological index includes the degree and clustering coefficient of each data object.

[0037] The method of subdividing each multi-dimensional data group into a plurality of segments includes:

[0038] The degree threshold and clustering coefficient threshold are preset, and data objects with a degree lower than the degree threshold are marked as potential outliers, and data objects with a clustering coefficient lower than the clustering coefficient threshold are marked as potential outliers; potential outliers are used as segmentation boundaries; the boundaries of the periodic components of the data time series are used as segmentation boundaries; the segmentation boundaries are the dividing lines between different segments, that is, several segments are obtained.

[0039] Furthermore, the calculation method of the short circuit risk index and the harmonic content index includes:

[0040] The standardized data corresponding to the insulation resistance and leakage current in each section of processed power data are obtained; each section of processed environmental data and the extracted standardized data corresponding to the insulation resistance and leakage current are input into a pre-built neural network-based assessment model, and a short circuit risk index is output, where the value range of the short circuit risk index is between 0 and 1;

[0041] The voltage and current in the processed power data are formed into segmented waveforms, namely, voltage waveforms and current waveforms; the voltage waveforms and current waveforms are subjected to fast Fourier transform to obtain spectrum components; and based on the spectrum components, the ratio of the effective value of the fundamental component to the effective value of all components is calculated, namely, the total harmonic content rate; the fundamental component is a special component in the spectrum components, which corresponds to the fundamental frequency of the waveform, namely, the lowest frequency component;

[0042] Collect voltage and current waveforms under different load types and working environments within a fixed historical time period; manually mark the expected harmonic impact level corresponding to each set of voltage and current waveforms;

[0043] The total harmonic content rate is extracted from the voltage waveform and the current waveform, and the relevant feature vectors of the environmental data corresponding to the load type and the working environment are extracted at the same time, which are recorded as the load type feature vector and the working environment feature vector; the total harmonic content rate, the load type feature vector, and the working environment feature vector are taken as input, and the manually marked expected harmonic impact level is taken as the output target, and a machine learning model is used for training to obtain a mapping function f; the machine learning model is a neural network, a decision tree, or a support vector machine; the output of the mapping function f is a harmonic content index in the range of 0-1;

[0044] When the load type is a nonlinear load, during the training process, the loss weight of high-order harmonics is increased, and based on the environmental data of the working environment, it is judged whether the working environment is a harsh working environment; the various types of environmental data are weighted and calculated to obtain the environmental value, and the environmental threshold is preset. The working environment corresponding to the environmental data with an environmental value greater than the environmental threshold is recorded as a harsh working environment; for harsh working environments, during the training process, the threshold value of the expected harmonic impact level is lowered.

[0045] Furthermore, the method of constructing the multi-objective optimization model includes:

[0046] The multi-objective optimization model is defined to consist of the optimization objective function FD(a) and constraints;

[0047] Optimization objective function FD(a)=min[f1(a),f2(a),f3(a)]; wherein, f1(a) represents the short-circuit risk function, and the calculated short-circuit risk index is used as the function value of the short-circuit risk function; f2(a) represents the harmonic content function, and the calculated harmonic content index is used as the function value of the harmonic content function; f3(a) represents the energy consumption function; a is the decision variable vector; the decision variable vector includes insulation level, filter parameters and reactive power compensation; the filter parameters include filter resistance and filter inductance;

[0048]

[0049] Among them, k1, k2 and k3 are the corresponding item weights; U is the operating voltage; Rins(a) is the insulation resistance; δ(E) is the environmental correction coefficient, E is the environmental data; I is the effective value of current, R(a) is the filter resistance, ω is the angular frequency; L(a) is the filter inductance, Q(a) is the reactive power compensation; H(M) is the compensation mode correction coefficient, M is the compensation type;

[0050] The constraints are that the insulation level, filter parameters and reactive power compensation are all within the preset corresponding ranges.

[0051] Furthermore, the method for solving the optimal setting of power parameters includes:

[0052] The optimization objective function is recombined into a weighted sum of a single objective function to obtain a single objective optimization function FH(a)=p1×f1(a)+p2×f2(a)+p3×f3(a); where p1, p2 and p3 are the corresponding objective weights;

[0053] Randomly generate m1 initial solutions in the solution space to form an initial population, where m1 is a positive integer greater than 1; each solution corresponds to a mass point; it contains a decision variable vector (insulation level, filter parameters and reactive compensation amount); define a gravity function, define the gravity function as a single-objective optimization function, and calculate the function value of the gravity function for each mass point as the mass of the corresponding mass point;

[0054] Iteration is performed based on the mass of the particle. The iteration process includes:

[0055] The preset black hole threshold is used to treat particles with a mass less than the black hole threshold as black holes; the gravitational force FL(l,j) between any two particles l and j in the initial population is calculated;

[0056] Where GP(t) is the gravitational constant of the t-th iteration, R(l,j) is the Euclidean distance between two particles l and j, τ is a constant; m(l) is the mass of particle l, m(j) is the mass of particle j; β1 and β2 are mass exponents, and γ1 is the distance exponent;

[0057] Where GP_in is the initial preset gravitational constant, GP_nal is the preset final gravitational constant, and Max_iter is the preset maximum number of iterations;

[0058] When there is a black hole particle among any two particles l and j involved in the calculation, the gravitational constant is replaced by the black hole gravitational constant GBH when calculating the gravitational force FL(l,j); the black hole gravitational constant GBH is initially set to several times the gravitational constant; when iterating, Among them, α1 is the adjustment coefficient, m_bh is the mass of the black hole;

[0059] In each iteration, a fixed proportion of mass points are randomly selected as thrust objects. For each thrust object J, the magnitude of the thrust acting on it is defined as Wherein, m(J) is the mass of the thrust object J; for each thrust object J, the thrust direction acting on it is defined as a random direction; the calculated thrust is superimposed with the gravity; and the resultant force FU(J) after superimposed thrust is obtained;

[0060] Calculate the acceleration of the thrust object J based on the resultant force FU(J) The direction of acceleration a(J) is the direction of the resultant force; for a particle that is not the object of thrust, its acceleration is the gravitational force divided by the corresponding mass;

[0061] Update the velocity and position of each particle according to the calculated acceleration;

[0062] The speed update formula is:

[0063] v(j,t+1)=rand_j×v(j,t)+a(j,t), where rand_j is a random number in the interval [0,1], v(j,t+1) is the velocity of particle j at the t+1th iteration, v(j,t) is the velocity of particle j at the tth iteration, and a(j,t) is the acceleration of particle j at the tth iteration;

[0064] The update formula of position is:

[0065] X(j,t+1)=X(j,t)+v(j,t+1); where X(j,t+1) is the position of particle j at the t+1th iteration, and X(j,t) is the position of particle j at the tth iteration;

[0066] Check whether the updated position exceeds the value range of the decision variable vector. If so, project the corresponding particle back to the solution space. When the preset maximum number of iterations Max_iter is reached, output the particle with the highest quality at this time as the optimal solution. The combination of insulation level, filter parameters and reactive power compensation corresponding to the optimal solution is the optimal setting of power parameters.

[0067] The technical effects and advantages of the safe power system based on short circuit and harmonic control in humid environment of the present invention are as follows:

[0068] The present invention can comprehensively improve the safety of electricity use and the quality of electric energy, greatly reduce the risk of short circuits and harmonic interference, has excellent environmental adaptability, and can maintain stable and reliable operation in various complex and changeable environments; through intelligent data processing and analysis, it can more accurately capture the essential characteristics and changing trends of data, and segmentation greatly improves the accuracy and efficiency of data analysis, so that the system can more keenly identify potential risks and anomalies, and can more deeply explore the value of data, and provide more accurate risk assessment and decision support; it can perform differentiated processing on data of different time periods and spatial characteristics, so as to provide more personalized and accurate safe electricity use solutions; its multi-objective optimization capability enables the system to achieve the best balance between safety, power quality and energy efficiency, and can automatically adjust parameters as the environment changes, always maintaining the best performance, which can not only significantly improve the safety and reliability of electricity use, but also reduce energy consumption, reduce economic losses, and create greater economic and social benefits for users. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 A schematic diagram of a safe power system based on short circuit and harmonic control in a humid environment according to the present invention;

[0070] Figure 2 It is a schematic diagram of a safe electricity use method based on short circuit and harmonic control in a humid environment according to the present invention. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] Example 1

[0073] See also Figure 1As shown, the safe electricity system based on short circuit and harmonic control in humid environments described in this embodiment includes: a data acquisition and processing module, an exponential fitting module and a strategy optimization module; each module is connected by wired and / or wireless means to realize data transmission between modules.

[0074] The data acquisition and processing module includes an acquisition unit and a processing unit; the acquisition unit is used to acquire environmental data and power data of the power system, and the processing unit is used to pre-process the environmental data and power data to obtain several segments of processed environmental data and processed power data;

[0075] The index fitting module is used to calculate several short circuit risk indexes and harmonic content indexes based on the processed environmental data and the processed power data;

[0076] The strategy optimization module includes a model construction unit and an optimization solution unit; the model construction unit is used to construct a multi-objective optimization model according to the short-circuit risk index and the harmonic content index; the optimization solution unit solves several corresponding optimal setting power parameters based on the constructed multi-objective optimization model; and the solved optimal setting power parameters are applied to the power system.

[0077] Furthermore, the environmental data include ambient temperature, ambient humidity and ambient dust concentration (dust in the environment may be deposited on the insulating parts of electrical equipment, reducing insulation performance and increasing the risk of flashover or insulation breakdown of the equipment); for seaside environments, the environmental data also includes ambient salt spray concentration; power data include voltage, current, insulation resistance, leakage current and load type; load types include resistive loads (such as electric heaters, heaters, lighting fixtures), inductive loads (equipment containing windings), non-linear loads (equipment containing power electronic devices), etc.

[0078] It should be noted that the environmental dust concentration is obtained using a laser scattering dust monitor, the insulation resistance is obtained using an insulation resistance tester, and the leakage current is obtained using a clamp-type leakage current sensor.

[0079] Furthermore, the environmental data and power data are pre-processed in the following ways:

[0080] The environmental data and power data are sorted into several time series in chronological order, and each time series contains all the environmental data and power data at a specific time point; several time series are combined side by side by columns to construct a two-dimensional data matrix, in which the rows of the two-dimensional data matrix represent time points, and the columns represent different environmental data and power data; a column is added to the two-dimensional data matrix to represent the timestamp corresponding to each row, and several columns are added to the two-dimensional data matrix to represent the spatial coordinates corresponding to each row of data, so as to obtain a multi-dimensional data matrix; the multi-dimensional data matrix is ​​standardized to eliminate the dimensional influence of data of different dimensions, so as to obtain a standard multi-dimensional data matrix.

[0081] A preliminary analysis is performed on the standard multi-dimensional data matrix to obtain several multi-dimensional data clusters; specifically, the standard multi-dimensional data matrix is ​​expanded into several data vectors by row, each data vector is regarded as a waveform signal, and several waveform data are constructed.

[0082] Perform waveform conversion on each waveform data to obtain the conversion domain coefficient of each waveform data; specifically, convolve the waveform data with the Coifl et wavelet basis function at different scales to obtain the wavelet decomposition coefficients of the corresponding scales; quantize the wavelet decomposition coefficients to obtain quantized wavelet decomposition coefficients, and pack all the quantized wavelet decomposition coefficients in a packing order to obtain the conversion domain coefficients; the packing order is level-by-level or tree-based.

[0083] Ways to quantize wavelet decomposition coefficients include:

[0084] Define the sparse representation of wavelet decomposition coefficients, that is, wavelet decomposition coefficients in, is the wavelet basis matrix (a matrix form of a set of basis functions used for wavelet transform), α is the sparse coefficient vector to be determined, and since most wavelet decomposition coefficients are close to zero, α is a sparse coefficient vector; the sparse representation is reconstructed to obtain the measurement process function, and the formula of the measurement process function is: Among them, θ is the measurement matrix (the matrix used to measure or observe the original signal), and y is the observation value (the compressed observation data obtained by measuring the wavelet decomposition coefficients through the measurement matrix).

[0085] Define the optimization model including objective function FH and constraint function;

[0086] FH=minΣw_i×|α(i)|0; the formula of constraint function is: Where w_i is the weight of the i-th sparse coefficient vector, α(i) is the i-th sparse coefficient vector; |α(i)|0 is the L0 norm of α(i).

[0087] Solve the optimization model to obtain the optimal sparse coefficients; retain the non-zero elements in the optimal sparse coefficients, and quantize the remaining elements to 0 to obtain a quantized sparse coefficient vector as the quantized wavelet decomposition coefficients.

[0088] Ways to solve the optimization model include:

[0089] Initialize a sparse solution α_0, which takes a vector of all zeros or a random vector; calculate the initial residual For the kth iteration, k is a positive integer greater than 1; calculate the gradient vector in, is the transpose of the wavelet basis matrix, θ T is the transpose of the measurement matrix; r_(k-1) is the residual of the k-1th iteration.

[0090] The gradient vector is thresholded to obtain an intermediate solution α'_k. The formula for thresholding is: α'_k=α_(k-1)+g_k; where α_(k-1) is the sparse solution obtained at the k-1th iteration.

[0091] Keep the k elements with the largest absolute values ​​in α'_k, set the rest to 0, and get a new sparse solution α_k (the sparse solution of the kth iteration); calculate a new residual based on the new sparse solution α_k The iterative process continues until the preset number of iterations is reached, and the last obtained sparse solution is output as the optimal sparse coefficient.

[0092] Eigenvalues ​​are extracted from the conversion domain coefficients, including energy, entropy and peak value; the extracted eigenvalues ​​are constructed into new eigenvectors in the feature space, and the eigenvectors corresponding to all data objects (each data point or data sample in the environmental data and power data) form a feature point cloud (composed of several feature points) in the feature space.

[0093] In the feature space, density estimation is performed on the neighborhood of each feature point (calculated using the eigenvector) to obtain the density estimation value corresponding to each feature point; the density estimation value corresponding to each feature point is normalized to obtain the corresponding density value; a density threshold is preset, and feature points with density values ​​greater than the density threshold are used as the center points of the initial multidimensional data cluster; for feature points that are not center points, they are assigned to the multidimensional data cluster where the nearest center point is located to obtain preliminary multidimensional data clusters; all preliminary multidimensional data clusters are post-processed to obtain several final multidimensional data clusters.

[0094] Post-processing methods include:

[0095] For each preliminary multi-dimensional data cluster, a corresponding weighted undirected graph G = (V, E) is constructed; V is a set of nodes, and E is a set of edges; each feature point in the preliminary multi-dimensional data cluster is used as a node of the weighted undirected graph G; the cosine similarity of any two feature points is calculated (based on the feature vector), and a similarity threshold is preset. If the cosine similarity is greater than the similarity threshold, an edge is connected between the nodes corresponding to the two feature points; the weight of the edge is the cosine similarity between the two feature points.

[0096] Calculate the degree matrix D and adjacency matrix A of the weighted undirected graph G corresponding to the preliminary multi-dimensional data group, and calculate the Laplacian matrix L=DA.

[0097] Calculate the vectors corresponding to the first K smallest non-zero eigenvalues ​​of the Laplace matrix L, map each feature point in the preliminary multidimensional data cluster to the K-dimensional space formed by these K vectors, and perform K-Means clustering on the feature points in the preliminary multidimensional data cluster in the K-dimensional space, that is, split the preliminary multidimensional data cluster into several subclusters, and all the obtained subclusters are multidimensional data clusters.

[0098] The time distribution characteristics and spatial distribution characteristics of the data objects within each multidimensional data group are obtained; specifically, for the data objects within each multidimensional data group, their timestamp information is extracted, and a data time series is constructed; for the data objects within each multidimensional data group, their spatial coordinate information is extracted; and the spatial distances between the data objects are calculated to construct a distance matrix.

[0099] The data time series is subjected to discrete Fourier transform (DFT) to obtain the amplitude spectrum in the frequency domain, and the power spectrum density is calculated based on the amplitude spectrum. Where N1 is the length of the data time series, and X(e) is the complex coefficient of the e-th frequency component obtained by discrete Fourier transform of the data time series, that is, the amplitude spectrum.

[0100] The calculated power spectrum density constitutes a power spectrum density graph. The peak value is found in the power spectrum density graph, and the frequency corresponding to the peak value is taken as the periodic component, which is the time distribution feature.

[0101] Take any two data objects in the distance matrix as data pairs, count the distances between all data pairs, and use them as weight parameters between data pairs.

[0102] Initialize each data object in the distance matrix as a tree containing only itself to form a forest F. Start from the data pair with the smallest weight parameter to traverse all data pairs. If the current data pair belongs to different trees in F, add this data pair to F and merge the two trees into one tree. If the current data pair belongs to the same tree in F, skip the data pair. Repeat until there is only one tree left in F. At this time, this tree is recorded as a greedy tree.

[0103] The topological index of the greedy tree is calculated as the spatial distribution feature. The topological index includes the degree of each data object (the number of connected edges) and the clustering coefficient (the connection degree value between neighboring data objects).

[0104] According to the time distribution characteristics and spatial distribution characteristics, each multidimensional data cluster is subdivided into several segments; specifically, a degree threshold and a clustering coefficient threshold are preset (set by multiple experimental fittings), and data objects with a degree lower than the degree threshold are marked as potential outliers, and data objects with a clustering coefficient lower than the clustering coefficient threshold are marked as potential outliers; potential outliers are used as segment boundaries; the boundaries of the periodic components of the data time series are used as segment boundaries; the segment boundaries are the dividing lines between different segments, that is, several segments are obtained.

[0105] According to the determined segment boundaries, the data objects of each segment are extracted from the standard multi-dimensional data matrix; and several segments of processed environmental data and several groups of processed power data are obtained.

[0106] Segmentation can better capture the essential characteristics of data and provide high-quality segmented data input for subsequent risk assessment and optimization decisions.

[0107] Furthermore, the calculation methods of the short circuit risk index and the harmonic content index include:

[0108] The standardized processed data corresponding to the insulation resistance and leakage current in each section of processed power data are obtained; each section of processed environmental data and the extracted standardized processed data corresponding to the insulation resistance and leakage current are input into a pre-built neural network-based assessment model, and a short-circuit risk index is output. The value range of the short-circuit risk index is between 0 and 1, and the larger the value, the higher the short-circuit risk.

[0109] The evaluation model is constructed by:

[0110] An artificial neural network is used as the skeleton structure of the evaluation model. The artificial neural network includes an input layer, a hidden layer and an output layer. The input layer has 5 neurons, corresponding to the standardized data corresponding to the insulation resistance and leakage current and the data type corresponding to the processed environmental data respectively. The hidden layer has 10 neurons, and the activation function adopts the Sigmoid function. The output layer has 1 neuron, corresponding to the short-circuit risk index Risk, and the value range is [0, 1]. The collected training data includes known environmental data, power data and corresponding short-circuit risk labels. The back-propagation algorithm is used to train the artificial neural network to minimize the mean square error between the output and the short-circuit risk label. The constructed evaluation model is obtained.

[0111] The voltage and current in the processed power data form segmented waveforms, namely, voltage waveforms and current waveforms. Fast Fourier transform (FFT) is performed on the voltage waveform and the current waveform to obtain spectrum components. Based on the spectrum components, the ratio of the effective value of the fundamental component to the effective value of all components is calculated, namely, the total harmonic content rate. The fundamental component is a special component in the spectrum components, which corresponds to the basic frequency of the waveform, namely, the lowest frequency component.

[0112] At a fixed historical time, voltage and current waveforms are collected under different load types (resistive, inductive, nonlinear) and working environments (temperature and humidity, dust, salt spray, etc.); the expected harmonic impact level (such as level 1-5) corresponding to each set of voltage and current waveforms is manually labeled.

[0113] The total harmonic content rate is extracted from the voltage waveform and the current waveform, and the relevant feature vectors of the environmental data corresponding to the load type and the working environment are extracted at the same time, which are recorded as the load type feature vector and the working environment feature vector.

[0114] The total harmonic content rate, load type feature vector, and working environment feature vector are taken as input, and the manually labeled expected harmonic impact level is taken as the output target. The mapping function f is obtained by training with a machine learning model; the machine learning model is a neural network, a decision tree, or a support vector machine; the output of the mapping function f is the harmonic content index in the range of 0-1.

[0115] When the load type is a nonlinear load, during the training process, the loss weight of high-order harmonics is increased to make the mapping function more sensitive to high-order harmonics; based on the environmental data of the working environment, it is judged whether the working environment is a harsh working environment; specifically, each type of environmental data is weighted and calculated to obtain an environmental value, and an environmental threshold is preset, and the working environment corresponding to the environmental data with an environmental value greater than the environmental threshold is recorded as a harsh working environment.

[0116] For harsh working environments, during the training process, the threshold value of the expected harmonic impact level (the dividing value used when mapping the total harmonic content rate to the harmonic impact level) is lowered. The reduction strategy is to reduce the multiple, and the multiple is determined according to the actual situation, so that the mapping function can output a higher harmonic content index at the same total harmonic content rate.

[0117] Ways to build a multi-objective optimization model include:

[0118] The multi-objective optimization model is defined to consist of the optimization objective function FD(a) and constraints;

[0119] The optimization objective function FD(a)=min[f1(a),f2(a),f3(a)]; wherein, f1(a) represents the short-circuit risk function, and the calculated short-circuit risk index is used as the function value of the short-circuit risk function; f2(a) represents the harmonic content function, and the calculated harmonic content index is used as the function value of the harmonic content function; f3(a) represents the energy consumption function; a is the decision variable vector; the decision variable vector includes insulation level, filter parameters and reactive power compensation; the filter parameters include filter resistance and filter inductance.

[0120]

[0121] Among them, k1, k2 and k3 are the corresponding item weights, all of which are constants; U is the operating voltage; Rins(a) is the insulation resistance, which depends on the insulation level; δ(E) is the environmental correction coefficient, and E is the environmental data; I is the effective value of the current, R(a) is the filter resistance, and ω is the angular frequency; L(a) is the filter inductance, Q(a) is the reactive power compensation amount; H(M) is the compensation method correction coefficient, and M is the compensation type (such as capacitor compensation, reactive power shunt compensator, synchronous motor reactive compensation, reactive generator compensation, etc.).

[0122] The environmental correction coefficient δ(E) is obtained by weighted summing the various types of environmental data using an exponential function. The exponential function can better characterize the nonlinear effect of environmental factors on insulation loss.

[0123] The compensation method correction coefficient H(M) is obtained by weighted summing the indicator functions of different compensation methods. The indicator function takes a value of 1 when the condition is met, and takes a value of 0 otherwise.

[0124] The constraints are that the insulation level, filter parameters and reactive power compensation are all within the preset corresponding ranges.

[0125] Furthermore, the solution for optimally setting power parameters includes:

[0126] The optimization objective function is recombined into the weighted sum of a single objective function to obtain the single objective optimization function FH(a)=p1×f1(a)+p2×f2(a)+p3×f3(a); where p1, p2 and p3 are the corresponding objective weights; f1(a), f2(a) and f3(a) are taken as single objective functions.

[0127] In the solution space (the data space where the insulation level, filter parameters and reactive power compensation amount are located), m1 initial solutions are randomly generated to form an initial population, where m1 is a positive integer greater than 1.

[0128] Each solution corresponds to a mass point; it contains a decision variable vector (insulation level, filter parameters and reactive compensation amount); defines a gravity function, defines the gravity function as a single-objective optimization function, and calculates the function value of the gravity function for each mass point as the mass of the corresponding mass point.

[0129] Iteration is performed based on the mass of the particle. The iteration process includes:

[0130] The preset black hole threshold treats particles with a mass less than the black hole threshold as black holes; calculates the gravitational force FL(l,j) between any two particles l and j in the initial population.

[0131] Among them, GP(t) is the gravitational constant of the tth iteration, R(l,j) is the Euclidean distance between two mass points l and j, τ is a very small constant to avoid the denominator being 0; m(l) is the mass of mass point l, m(j) is the mass of mass point j; β1 and β2 are mass exponents used to adjust the influence of mass on gravity, and γ1 is the distance exponent used to adjust the influence of distance on gravity.

[0132] Where GP_in is the initial preset gravitational constant, GP_nal is the preset final gravitational constant, and Max_iter is the preset maximum number of iterations.

[0133] When any two particles l and j involved in the calculation have a particle that is a black hole, the gravitational constant is replaced by the black hole gravitational constant GBH when calculating the gravitational force FL(l,j); the black hole gravitational constant GBH is initially set to several times the gravitational constant, for example, 10 times; when iterating, Among them, α1 is the adjustment coefficient and m_bh is the mass of the black hole.

[0134] In each iteration, a fixed proportion of mass points are randomly selected as thrust objects. For each thrust object J, the magnitude of the thrust acting on it is defined as Wherein, m(J) is the mass of the thrust object J. For each thrust object J, the thrust direction acting on it is defined as a random direction. The calculated thrust is superimposed on the gravity to obtain the resultant force FU(J) after superimposed thrust.

[0135] Calculate the acceleration of the thrust object J based on the resultant force FU(J) The direction of acceleration a(J) is the direction of the resultant force; for a particle that is not the object of thrust, its acceleration is the gravitational force divided by the corresponding mass;

[0136] Update the velocity and position of each particle based on the calculated acceleration.

[0137] The speed update formula is:

[0138] v(j,t+1)=rand_j×v(j,t)+a(j,t), where rand_j is a random number in the interval [0,1], v(j,t+1) is the velocity of particle j at the t+1th iteration, v(j,t) is the velocity of particle j at the tth iteration, and a(j,t) is the acceleration of particle j at the tth iteration;

[0139] The update formula of position is:

[0140] X(j,t+1)=X(j,t)+v(j,t+1); where X(j,t+1) is the position of particle j at the t+1th iteration, and X(j,t) is the position of particle j at the tth iteration.

[0141] Check whether the updated position exceeds the value range of the decision variable vector. If so, project the corresponding particle back to the solution space. When the preset maximum number of iterations Max_iter is reached, output the particle with the highest quality at this time as the optimal solution. The combination of insulation level, filter parameters and reactive power compensation corresponding to the optimal solution is the optimal setting of power parameters.

[0142] This embodiment can comprehensively improve the safety and quality of electricity use, significantly reduce the risk of short circuits and harmonic interference, has excellent environmental adaptability, and can maintain stable and reliable operation in various complex and changeable environments; through intelligent data processing and analysis, it can more accurately capture the essential characteristics and changing trends of data, and segmentation greatly improves the accuracy and efficiency of data analysis, so that the system can more keenly identify potential risks and anomalies, and can more deeply explore the value of data, and provide more accurate risk assessment and decision support; it performs differentiated processing on data of different time periods and spatial characteristics, thereby providing more personalized and accurate safe electricity use solutions; its multi-objective optimization capability enables the system to achieve the best balance between safety, power quality and energy efficiency, and can automatically adjust parameters as the environment changes to always maintain optimal performance, which can not only significantly improve the safety and reliability of electricity use, but also reduce energy consumption, reduce economic losses, and create greater economic and social benefits for users.

[0143] Example 2

[0144] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides a safe electricity use method based on short circuit and harmonic control in a humid environment, including:

[0145] S1, collecting environmental data and power data of the power system, pre-processing the environmental data and power data to obtain several sections of processed environmental data and processed power data;

[0146] S2. Calculate several short circuit risk indexes and harmonic content indexes based on the processed environmental data and the processed power data;

[0147] S3. According to the short-circuit risk index and the harmonic content index, a multi-objective optimization model is constructed. Based on the constructed multi-objective optimization model, several corresponding optimal power setting parameters are solved; the solved optimal power setting parameters are applied to the power system to achieve safe electricity use.

[0148] Example 3

[0149] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the above-mentioned method for safe electricity use based on short circuit and harmonic control in a humid environment is implemented.

[0150] Since the electronic device introduced in this embodiment is an electronic device used to implement a safe method for using electricity based on short circuit and harmonic control in a humid environment in the embodiment of the present application, based on the safe method for using electricity based on short circuit and harmonic control in a humid environment introduced in the embodiment of the present application, a person skilled in the art can understand the specific implementation method of the electronic device of the present embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present application is not described in detail here. As long as a person skilled in the art implements an electronic device used in a safe method for using electricity based on short circuit and harmonic control in a humid environment in the embodiment of the present application, it belongs to the scope of protection of the present application.

[0151] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0152] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A safe power system based on short circuit and harmonic control in humid environment, characterized in that: include: Data acquisition and processing module, index fitting module and strategy optimization module; Each module is connected via wired and / or wireless means; The data acquisition and processing module includes an acquisition unit and a processing unit; the acquisition unit is used to acquire environmental data and power data of the power system, and the processing unit is used to pre-process the environmental data and power data to obtain several segments of processed environmental data and processed power data; The index fitting module is used to calculate several short circuit risk indexes and harmonic content indexes based on the processed environmental data and the processed power data; The strategy optimization module includes a model construction unit and an optimization solution unit; the model construction unit is used to construct a multi-objective optimization model according to the short-circuit risk index and the harmonic content index; the optimization solution unit solves several corresponding optimal setting power parameters based on the constructed multi-objective optimization model; and the solved optimal setting power parameters are applied to the power system.

2. A safe power system based on short circuit and harmonic control in humid environment according to claim 1, characterized in that: The environmental data includes the ambient temperature, ambient humidity and ambient dust concentration; for the seaside environment, the environmental data also includes the ambient salt spray concentration; the power data includes voltage, current, insulation resistance, leakage current and load type.

3. A safe power system based on short circuit and harmonic control in humid environment according to claim 2, characterized in that: The method of preprocessing the environmental data and the power data includes: The environmental data and the power data are sorted into several time series in chronological order, and each time series contains all the environmental data and power data at a specific time point; several time series are combined side by side by columns to construct a two-dimensional data matrix, and the rows of the two-dimensional data matrix represent time points, and the columns represent different environmental data and power data; a column is added to the two-dimensional data matrix to represent the timestamp corresponding to each row, and several columns are added to the two-dimensional data matrix to represent the spatial coordinates corresponding to each row of data, so as to obtain a multi-dimensional data matrix; the multi-dimensional data matrix is ​​standardized to obtain a standard multi-dimensional data matrix; the standard multi-dimensional data matrix is ​​preliminarily analyzed to obtain several multi-dimensional data groups; the time distribution characteristics and spatial distribution characteristics of the data objects within each multi-dimensional data group are obtained; according to the time distribution characteristics and spatial distribution characteristics, each multi-dimensional data group is subdivided into several segments; according to the determined segment boundaries, the data objects of each segment are extracted from the standard multi-dimensional data matrix; several segments of processed environmental data and several groups of processed power data are obtained.

4. A safe power system based on short circuit and harmonic control in humid environment according to claim 3, characterized in that: The method of performing preliminary analysis on the standard multi-dimensional data matrix includes: Expand the standard multi-dimensional data matrix into several data vectors by row, regard each data vector as a waveform signal, and construct several waveform data; convolve the waveform data with the Coiflet wavelet basis function at different scales to obtain the wavelet decomposition coefficients of the corresponding scales; quantize the wavelet decomposition coefficients to obtain quantized wavelet decomposition coefficients, and pack all the quantized wavelet decomposition coefficients in a packing order to obtain conversion domain coefficients; Extract eigenvalues ​​from the conversion domain coefficients, including energy, entropy and peak value; construct the extracted eigenvalues ​​into new eigenvectors in the feature space, and the eigenvectors corresponding to all data objects form a feature point cloud in the feature space; In the feature space, density estimation is performed on the neighborhood of each feature point to obtain the density estimation value corresponding to each feature point; the density estimation value corresponding to each feature point is normalized to obtain the corresponding density value; a density threshold is preset, and the feature point with a density value greater than the density threshold is used as the center point of the initial multi-dimensional data cluster; for feature points that are not center points, they are assigned to the multi-dimensional data cluster where the nearest center point is located to obtain preliminary multi-dimensional data clusters; all preliminary multi-dimensional data clusters are post-processed to obtain several final multi-dimensional data clusters.

5. A safe power system based on short circuit and harmonic control in humid environment according to claim 4, characterized in that: The method of quantizing the wavelet decomposition coefficients includes: Define the sparse representation of wavelet decomposition coefficients, that is, wavelet decomposition coefficients in, is the wavelet basis matrix, α is the sparse coefficient vector to be determined; the sparse representation is reconstructed to obtain the measurement process function, and the formula of the measurement process function is: Among them, θ is the measurement matrix and y is the observation value; Define the optimization model including objective function FH and constraint function; FH=min∑w_i×|α(i)|0; the formula of constraint function is: Where w_i is the weight of the i-th sparse coefficient vector, α(i) is the i-th sparse coefficient vector; |α(i)|0 is the L0 norm of α(i); Solve the optimization model to obtain the optimal sparse coefficients; retain the non-zero elements in the optimal sparse coefficients, quantize the remaining elements to 0, and obtain the quantized sparse coefficient vector as the quantized wavelet decomposition coefficients; The method of solving the optimization model includes: Initialize a sparse solution α_0, which takes a vector of all zeros or a random vector; calculate the initial residual For the kth iteration; calculate the gradient vector in, is the transpose of the wavelet basis matrix, θ T is the transpose of the measurement matrix; r_(k-1) is the residual of the k-1th iteration; The gradient vector is thresholded to obtain the intermediate solution α'_k. The formula for thresholding is: α'_k=α_(k-1)+g_k; where α_(k-1) is the sparse solution obtained at the k-1th iteration; Keep the k elements with the largest absolute values ​​in α'_k, set the rest to 0, and get a new sparse solution α_k; calculate a new residual based on the new sparse solution α_k The iterative process continues until the preset number of iterations is reached, and the last obtained sparse solution is output as the optimal sparse coefficient.

6. A safe power system based on short circuit and harmonic control in humid environment according to claim 5, characterized in that: The method of post-processing all preliminary multi-dimensional data groups includes: For each preliminary multi-dimensional data cluster, a corresponding weighted undirected graph G = (V, E) is constructed; V is a set of nodes, and E is a set of edges; each feature point in the preliminary multi-dimensional data cluster is used as a node of the weighted undirected graph G; the cosine similarity of any two feature points is calculated, and a similarity threshold is preset. If the cosine similarity is greater than the similarity threshold, an edge is connected between the nodes corresponding to the two feature points; the weight of the edge is the cosine similarity between the two feature points; Calculate the degree matrix D and adjacency matrix A of the weighted undirected graph G corresponding to the preliminary multi-dimensional data group, and calculate the Laplacian matrix L = DA; Calculate the vectors corresponding to the first K smallest non-zero eigenvalues ​​of the Laplace matrix L, map each feature point in the preliminary multidimensional data cluster to the K-dimensional space formed by these K vectors, and perform K-Means clustering on the feature points in the preliminary multidimensional data cluster in the K-dimensional space, that is, split the preliminary multidimensional data cluster into several subclusters, and all the obtained subclusters are multidimensional data clusters.

7. A safe power system based on short circuit and harmonic control in humid environment according to claim 6, characterized in that: The acquisition method of the time distribution characteristics and the space distribution characteristics includes: For each data object in a multi-dimensional data group, extract its timestamp information and construct a data time series; for each data object in a multi-dimensional data group, extract its spatial coordinate information; and calculate the spatial distance between data objects to construct a distance matrix; Perform discrete Fourier transform on the data time series to obtain the amplitude spectrum in the frequency domain, and calculate the power spectrum density based on the amplitude spectrum. Where N1 is the length of the data time series, X(e) is the complex coefficient of the e-th frequency component obtained by discrete Fourier transform of the data time series, that is, the amplitude spectrum; The calculated power spectrum density constitutes a power spectrum density graph. The peak value is found in the power spectrum density graph, and the frequency corresponding to the peak value is taken as the periodic component, which is the time distribution feature. Take any two data objects in the distance matrix as data pairs, count the distances between all data pairs, and use them as weight parameters between data pairs; Initialize each data object in the distance matrix as a tree containing only itself to form a forest F. Start from the data pair with the smallest weight parameter and traverse all data pairs. If the current data pair belongs to different trees in F, add this data pair to F and merge the two trees into one tree. If the current data pair belongs to the same tree in F, skip the data pair. Repeat until there is only one tree left in F. At this time, this tree is recorded as a greedy tree. Calculate the topological index of the greedy tree as the spatial distribution feature. The topological index includes the degree and clustering coefficient of each data object. The method of subdividing each multi-dimensional data group into a plurality of segments includes: The degree threshold and clustering coefficient threshold are preset, and data objects with a degree lower than the degree threshold are marked as potential outliers, and data objects with a clustering coefficient lower than the clustering coefficient threshold are marked as potential outliers; potential outliers are used as segmentation boundaries; the boundaries of the periodic components of the data time series are used as segmentation boundaries; the segmentation boundaries are the dividing lines between different segments, that is, several segments are obtained.

8. A safe power system based on short circuit and harmonic control in humid environment according to claim 7, characterized in that: The calculation method of the short circuit risk index and the harmonic content index includes: The standardized data corresponding to the insulation resistance and leakage current in each section of processed power data are obtained; each section of processed environmental data and the extracted standardized data corresponding to the insulation resistance and leakage current are input into a pre-built neural network-based assessment model, and a short circuit risk index is output, where the value range of the short circuit risk index is between 0 and 1; The voltage and current in the processed power data are formed into segmented waveforms, namely, voltage waveforms and current waveforms; the voltage waveforms and current waveforms are subjected to fast Fourier transform to obtain spectrum components; and based on the spectrum components, the ratio of the effective value of the fundamental component to the effective value of all components is calculated, namely, the total harmonic content rate; the fundamental component is a special component in the spectrum components, which corresponds to the fundamental frequency of the waveform, namely, the lowest frequency component; Collect voltage and current waveforms under different load types and working environments within a fixed historical time period; manually mark the expected harmonic impact level corresponding to each set of voltage and current waveforms; The total harmonic content rate is extracted from the voltage waveform and the current waveform, and the relevant feature vectors of the environmental data corresponding to the load type and the working environment are extracted at the same time, which are recorded as the load type feature vector and the working environment feature vector; the total harmonic content rate, the load type feature vector, and the working environment feature vector are taken as input, and the manually marked expected harmonic impact level is taken as the output target, and a machine learning model is used for training to obtain a mapping function f; the machine learning model is a neural network, a decision tree, or a support vector machine; the output of the mapping function f is a harmonic content index in the range of 0-1; When the load type is a nonlinear load, during the training process, the loss weight of high-order harmonics is increased, and based on the environmental data of the working environment, it is judged whether the working environment is a harsh working environment; the various types of environmental data are weighted and calculated to obtain the environmental value, and the environmental threshold is preset. The working environment corresponding to the environmental data with an environmental value greater than the environmental threshold is recorded as a harsh working environment; for harsh working environments, during the training process, the threshold value of the expected harmonic impact level is lowered.

9. A safe power system based on short circuit and harmonic control in humid environment according to claim 8, characterized in that: The method of constructing the multi-objective optimization model includes: The multi-objective optimization model is defined to consist of the optimization objective function FD(a) and constraints; Optimization objective function FD(a)=min[f1(a),f2(a),f3(a)]; wherein, f1(a) represents the short-circuit risk function, and the calculated short-circuit risk index is used as the function value of the short-circuit risk function; f2(a) represents the harmonic content function, and the calculated harmonic content index is used as the function value of the harmonic content function; f3(a) represents the energy consumption function; a is the decision variable vector; the decision variable vector includes insulation level, filter parameters and reactive power compensation; the filter parameters include filter resistance and filter inductance; Among them, k1, k2 and k3 are the corresponding item weights; U is the operating voltage; Rins(a) is the insulation resistance; δ(E) is the environmental correction coefficient, E is the environmental data; I is the effective value of current, R(a) is the filter resistance, ω is the angular frequency; L(a) is the filter inductance, Q(a) is the reactive power compensation; H(M) is the compensation mode correction coefficient, M is the compensation type; The constraints are that the insulation level, filter parameters and reactive power compensation are all within the preset corresponding ranges.

10. A safe power system based on short circuit and harmonic control in humid environment according to claim 9, characterized in that: The method for solving the optimal setting of power parameters includes: The optimization objective function is recombined into a weighted sum of a single objective function to obtain a single objective optimization function FH(a)=p1×f1(a)+p2×f2(a)+p3×f3(a); where p1, p2 and p3 are the corresponding objective weights; Randomly generate m1 initial solutions in the solution space to form an initial population, where m1 is a positive integer greater than 1; each solution corresponds to a mass point; it contains a decision variable vector (insulation level, filter parameters and reactive compensation amount); define a gravity function, define the gravity function as a single-objective optimization function, and calculate the function value of the gravity function for each mass point as the mass of the corresponding mass point; Iteration is performed based on the mass of the particle. The iteration process includes: The preset black hole threshold is used to treat particles with a mass less than the black hole threshold as black holes; the gravitational force FL(l,j) between any two particles l and j in the initial population is calculated; Where GP(t) is the gravitational constant of the t-th iteration, R(l,j) is the Euclidean distance between two particles l and j, τ is a constant; m(l) is the mass of particle l, m(j) is the mass of particle j; β1 and β2 are mass exponents, and γ1 is the distance exponent; Where GP_in is the initial preset gravitational constant, GP_nal is the preset final gravitational constant, and Max_iter is the preset maximum number of iterations; When there is a black hole particle among any two particles l and j involved in the calculation, the gravitational constant is replaced by the black hole gravitational constant GBH when calculating the gravitational force FL(l,j); the black hole gravitational constant GBH is initially set to several times the gravitational constant; when iterating, Among them, α1 is the adjustment coefficient, m_bh is the mass of the black hole; In each iteration, a fixed proportion of mass points are randomly selected as thrust objects. For each thrust object J, the magnitude of the thrust acting on it is defined as Wherein, m(J) is the mass of the thrust object J; for each thrust object J, the thrust direction acting on it is defined as a random direction; the calculated thrust is superimposed with the gravity; and the resultant force FU(J) after superimposed thrust is obtained; Calculate the acceleration of the thrust object J based on the resultant force FU(J) The direction of acceleration a(J) is the direction of the resultant force; for a particle that is not the object of thrust, its acceleration is the gravitational force divided by the corresponding mass; Update the velocity and position of each particle according to the calculated acceleration; The speed update formula is: v(j,t+1)=rand_j×v(j,t)+a(j,t), where rand_j is a random number in the interval [0,1], v(j,t+1) is the velocity of particle j at the t+1th iteration, v(j,t) is the velocity of particle j at the tth iteration, and a(j,t) is the acceleration of particle j at the tth iteration; The update formula of position is: X(j,t+1)=X(j,t)+v(j,t+1); where X(j,t+1) is the position of particle j at the t+1th iteration, and X(j,t) is the position of particle j at the tth iteration; Check whether the updated position exceeds the value range of the decision variable vector. If so, project the corresponding particle back to the solution space. When the preset maximum number of iterations Max_iter is reached, output the particle with the highest quality at this time as the optimal solution. The combination of insulation level, filter parameters and reactive power compensation corresponding to the optimal solution is the optimal setting of power parameters.

Images