A remote monitoring system for fault information of low voltage distribution cabinet

Abstract

The present invention belongs to the technical field of fault monitoring, and discloses a remote monitoring system for fault information of a low-voltage power distribution cabinet; operating data of the low-voltage power distribution cabinet is collected, and the operating data is preprocessed to obtain a preprocessed data set; feature extraction is performed on the preprocessed data set to obtain a fault feature vector; the fault feature vector is optimized to obtain an optimized fault feature vector; based on the optimized fault feature vector, a fault type and a fault location are identified; the identified fault type and fault location are sent to a remote monitoring terminal to realize remote fault monitoring of the low-voltage power distribution cabinet; the fault processing time can be greatly shortened, and the economic loss caused by the fault can be reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of fault monitoring, and more specifically, to a remote monitoring system for fault information of a low-voltage power distribution cabinet. Background Art

[0002] At present, the low-voltage distribution system is the "last mile" of power supply, and its operating status is directly related to the quality and safety of users' electricity consumption. However, the existing low-voltage distribution cabinet fault monitoring technology has many shortcomings, which brings serious hidden dangers to the reliable operation of the distribution system. First, traditional fault monitoring mainly relies on manual regular inspections and on-site maintenance, which is inefficient and delayed. Due to the wide distribution range and large number of low-voltage distribution cabinets, huge human and material resources are invested, and it is difficult to achieve timely and comprehensive monitoring, which can easily lead to faults being ignored or delayed in processing, posing a serious threat to power supply reliability. Secondly, the existing fault diagnosis methods often only focus on single data features and cannot comprehensively analyze the various operating conditions of the distribution cabinet. data, resulting in one-sided and inaccurate diagnostic results; the distribution cabinet is a complex electrical system, its operating status is affected by many factors, a single feature is difficult to reflect the whole picture, it is likely to miss or misdiagnose; furthermore, the existing methods lack an effective optimization mechanism in feature extraction and fault identification, and cannot fully tap the potential characteristics of the data, affecting the accuracy of fault identification; with the continuous development of the distribution system, the fault form and characteristics are becoming more and more complex, and a single fault pattern matching method can no longer meet actual needs; in addition, most of the existing technologies are unable to locate the specific location of the fault, which brings great difficulties to the subsequent fault investigation and maintenance, resulting in a prolonged fault handling cycle, causing greater economic losses and social impacts.

[0003] In view of this, the present invention proposes a remote monitoring system for low-voltage distribution cabinet fault information to solve the above problem. Summary of the invention

[0004] 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 remote monitoring system for fault information of a low-voltage power distribution cabinet, comprising: a data acquisition and processing module, for collecting operating data of the low-voltage power distribution cabinet, and preprocessing the operating data to obtain a preprocessed data set;

[0005] A feature extraction module is used to extract features from the preprocessed data set to obtain a fault feature vector;

[0006] An optimization module, used for optimizing the fault feature vector to obtain an optimized fault feature vector;

[0007] The fault identification module identifies the fault type and fault location based on the optimized fault feature vector; sends the identified fault type and fault location to the remote monitoring terminal; and each module is connected by wire and / or wireless means.

[0008] Furthermore, the operating data includes voltage data, current data, power data, electric energy data, harmonic data, temperature data, switch status data and protection action data; the voltage data includes the voltage value of each phase and the line voltage value; the current data includes the current value of each phase and the neutral line current value; the power data includes active power, reactive power and apparent power; the electric energy data includes active electric energy and reactive electric energy; the harmonic data includes the content of each harmonic; the temperature data includes the temperature inside the cabinet and the equipment temperature; the switch status data includes the switch status of each fuse and contactor; the protection action data is the action signal of each protection device.

[0009] Furthermore, the method of preprocessing the operation data includes:

[0010] Arrange the operation data of the low-voltage distribution cabinet in chronological order according to the data type to obtain the corresponding data time series, and divide the data time series into several time windows, each of which contains a continuous data point; define a two-dimensional coordinate plane, and for each time window, draw a line graph of the data points on the two-dimensional coordinate plane; add a scan line to the line graph, and place the initial position of the scan line at the first data point; move the scan line step by step from left to right, passing through each data point, and in the process of gradually moving the scan line, record the number of intersections between the horizontal scan line and the line graph;

[0011] Calculate the trend score within the time window corresponding to the data time series based on the number of recorded intersections ;

[0012] ;in, is the change in the number of intersections between the scan line and the line graph under the preset number of moving steps, is the number of data points in the corresponding time window, is the slope of the best-fit straight line of the line graph, , and is the weight coefficient, is a regulation factor greater than 0; is a comprehensive determining factor;

[0013] Comprehensive determinants ;in, For data points on the line chart The vertical coordinate of For data points on the line chart The horizontal axis of is the intercept of the best-fit straight line of the line graph, is a tuning parameter greater than 0, is the slope weight factor;

[0014] A trend threshold is preset, and the time window with a trend score greater than the trend threshold is recorded as the trend time window. The number of all trend time windows in the statistical data time series is recorded as the trend number. A sequence threshold is preset, and the data type corresponding to the data time series with a trend number greater than the sequence threshold is taken as the data to be processed; the data time series corresponding to the data to be processed is taken as the node of the tree root to construct a segmentation tree; and a stationary data sequence is obtained; a comprehensive test is performed on the stationary data sequence to obtain the preprocessed data, and all the preprocessed data constitute the preprocessed data set.

[0015] Furthermore, the method of constructing the segmentation tree includes:

[0016] Define the cost function , calculate the cost function of the node at the root of the tree The function value of ; denoted as split value;

[0017] ;in, is the residual sum of squares of the corresponding node, is the number of leaf nodes in the current segmentation tree, that is, the number of segments, is the preset maximum number of segments, is the average segment length, and is the benchmark parameter, For the The length of the segment, is the penalty coefficient; is the length penalty coefficient; is the absolute value of the difference between adjacent data points of the corresponding node;

[0018] If the split value is greater than the preset split threshold, the root node is split into two to generate two child nodes, and this process is repeated until the function values ​​of the cost functions of all nodes are less than the split threshold.

[0019] Furthermore, the method of obtaining a stationary data sequence includes:

[0020] Traverse all leaf nodes in the segmentation tree. Each leaf node corresponds to a segment of the data time series. Record the start and end time points of each segment as the segment boundary. Extract the data points in the corresponding segment based on the segment boundary, and fit a straight line using the least squares method. Use the corresponding straight line to calculate the residual for each data point in the corresponding segment extracted. The obtained residuals constitute a residual sequence. Connect the residual sequences of all segments to obtain the stable data sequence corresponding to the data to be processed.

[0021] Furthermore, the method of performing comprehensive detection on the stationary data sequence includes:

[0022] Calculate the stationary data series Order autocorrelation coefficient ;

[0023] ;in, For a stationary data series The value at a time point, is the mean of the stationary data series, is the residual attenuation parameter; For a stationary data series The value at each time point;

[0024] based on Order autocorrelation coefficient , calculate the preliminary series summary statistics ;

[0025] ;in, is the length of the stationary data sequence; a stationary data sequence is defined as a sequence containing The model with parameters is generated, then the comprehensive statistics of the series are calculated ; Define the significance level B, and find the chi-square distribution table to find the degree of freedom The chi-square distribution critical value ;

[0026] like Less than , the corresponding stationary data sequence is recorded as the preprocessed data; if Greater than or equal to , then adjust the cost function The parameters within the partition tree are rebuilt, and the stationary data sequence corresponding to the data to be processed is obtained again; until the calculated Less than .

[0027] Furthermore, the method of extracting features from the preprocessed data set includes:

[0028] Map the preprocessed data set to a high-dimensional space to obtain multidimensional features corresponding to the time step, where each dimension in the multidimensional features corresponds to a feature; all multidimensional features constitute a multidimensional feature set;

[0029] Divide the multidimensional feature set into a training set and a test set; perform grid processing on the training set data, define the number of grids M2 and the clock period T1; divide the multidimensional features in the training set and the test set into a number of clock windows of length T1 in the order of time steps; discretize the multidimensional features in each clock window, map the value of the multidimensional feature to the index of the discrete grid, and count the number of times each grid appears in the clock window to obtain a grid count vector;

[0030] Set a grid threshold, obtain its grid count vector for each clock window, and traverse each element v_j in the grid count vector, where j is the index of the element; if v_j is greater than or equal to the grid threshold, quantize the element v_j to 1; if v_j is less than the grid threshold, quantize the element v_j to 0; then obtain the quantized grid count vector, which is the grid pattern of the clock window, and count the grid patterns of all time windows to obtain a grid pattern set;

[0031] Build a clock model and define the number of states , take the size of the grid pattern set, initialize The state transition probability matrix A, in which the element A(u, r) represents the state transition probability from state u to state r, and the initial value is set to ;initialization The emission probability matrix C, where C(r, p) represents the emission probability of emitting to grid p in state r, and the initial value is set to ;

[0032] Use the training set to estimate the parameters of the clock model, and use the expectation maximization algorithm to estimate the state transition probability and the emission probability until the preset maximum number of iterations is reached; and obtain the trained clock model;

[0033] Connect the grid count vectors of the test set in all time windows to form a grid count sequence O. Based on the grid count sequence O, use the Viterbi algorithm on the trained clock model to calculate the most likely state sequence, and use the most likely state sequence as the fault feature vector.

[0034] Furthermore, the method of optimizing the fault feature vector includes:

[0035] Initialize optimization parameters, including population size , Iterations , propagation constant , Water wave height control parameters , Water wave length control parameters ;

[0036] Encode the fault feature vector into a binary string and randomly generate binary individuals constitute the initial population;

[0037] For each binary individual, decode it into a corresponding fault feature vector, and use the decoded fault feature vector as historical data to construct a historical data set, which includes several fault feature vectors and labels corresponding to the fault feature vectors; the label is the state of the low-voltage distribution cabinet; divide the historical data set into a training data set and a verification data set;

[0038] Use the classifier to train on the training data set to obtain the classification model, test the classification model on the validation data set, and calculate the classification accuracy as the water wave value corresponding to the binary individual; iteratively update each binary individual until the number of iterations is reached , output the binary individual corresponding to the optimal solution, and decode the corresponding binary individual to obtain the optimized fault feature vector.

[0039] Furthermore, the method of iteratively updating each binary individual includes:

[0040] Sort all corresponding binary individuals according to the water wave value from small to large. binary individuals, calculate their water wave values ; Based on its water wave value Calculate the propagation distance ;

[0041] ;in, A constant, is the current iteration number, is the influence constant; The maximum difference in water wave values ​​between binary individuals in the population;

[0042] Based on the calculated propagation distance , generating new binary individual solutions ;

[0043] ;in, To represent the first The current solution of binary individuals, is the solution for the binary individual with the highest water wave value in the current population; For one dimensional standard normally distributed random vector, is the dimension of the solution;

[0044] right Perform boundary processing and calculate The corresponding binary individual water wave value new_fit; if new_fit is greater than , then update the The solution for binary individuals is , otherwise, calculate the water wave height ;in, is a normal number, is a transition parameter between 0 and 1;

[0045] Based on water wave height , generating another new binary individual solution ;in, is a random number between 0 and 1; Perform boundary processing; calculate The corresponding binary individual water wave value new_fot, if new_fot is greater than , then update the The solution for binary individuals is , and record the current optimal solution and its water wave value.

[0046] Furthermore, the method for identifying the fault type and the fault location includes:

[0047] Construct a fault type library and a fault location library; the fault type library contains various fault types that may occur in the low-voltage distribution cabinet; the fault location library contains the location information of each component in the low-voltage distribution cabinet, recorded as the fault location; decode the optimized fault feature vector to obtain the corresponding fault feature; initialize the fault type of the low-voltage distribution cabinet to be empty and initialize the fault location of the low-voltage distribution cabinet to be empty; traverse each fault type in the fault type library, calculate the similarity between the fault type and the extracted fault feature, and select the fault type with the greatest similarity to the fault feature as the identified fault type;

[0048] Traverse each fault location in the fault location library, calculate the probability of the current fault location having the fault type according to the identified fault type, and select the fault location with the highest probability as the identification result.

[0049] The technical effects and advantages of the remote monitoring system for low-voltage power distribution cabinet fault information of the present invention are as follows:

[0050] The present invention realizes real-time remote monitoring of low-voltage distribution cabinet faults, greatly improves the timeliness and work efficiency of fault monitoring, helps to discover hidden faults as soon as possible, avoid the expansion of faults, and ensure the reliable operation of the distribution system; secondly, the system adopts a multi-source heterogeneous data fusion analysis method, which can comprehensively acquire and analyze various operating data such as voltage, current, power, temperature, etc. of the distribution cabinet, fully reflect the actual operating status of the distribution cabinet, provide rich data support for fault diagnosis, and greatly improve the accuracy and reliability of fault identification; thirdly, the system can automatically mine the fault features hidden in the data, overcome the defects of traditional methods that rely on manual experience, improve the effectiveness of fault feature extraction and optimization, thereby further improving the accuracy of fault identification; in addition, it can not only identify the fault type, but also locate the specific location where the fault occurs, provide direct guidance for subsequent fault investigation and maintenance, and can greatly shorten the fault handling time and reduce the economic losses caused by the fault. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of a remote monitoring system for fault information of a low-voltage power distribution cabinet according to the present invention;

[0052] Figure 2 A schematic diagram of a remote monitoring method for fault information of a low-voltage power distribution cabinet according to the present invention. DETAILED DESCRIPTION

[0053] 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.

[0054] Example 1

[0055] See also Figure 1 As shown, the remote monitoring system for low-voltage power distribution cabinet fault information described in this embodiment includes:

[0056] The data acquisition and processing module is used to collect the operation data of the low-voltage distribution cabinet and pre-process the operation data to obtain a pre-processed data set;

[0057] A feature extraction module is used to extract features from the preprocessed data set to obtain a fault feature vector;

[0058] An optimization module, used for optimizing the fault feature vector to obtain an optimized fault feature vector;

[0059] The fault identification module identifies the fault type and fault location based on the optimized fault feature vector; sends the identified fault type and fault location to the remote monitoring terminal to realize remote fault monitoring of the low-voltage distribution cabinet; and each module is connected by wired and / or wireless means to realize data transmission between modules.

[0060] The operating data includes voltage data, current data, power data, electric energy data, harmonic data, temperature data, switch status data and protection action data; the voltage data includes the voltage value of each phase and the line voltage value, reflecting the power supply situation of the distribution cabinet; the current data includes the current value of each phase and the neutral line current value, reflecting the load situation; the power data includes active power, reactive power and apparent power, reflecting the load characteristics; the electric energy data includes active electric energy and reactive electric energy, reflecting the energy consumption situation; the harmonic data includes the content of each harmonic, reflecting the power quality situation; the temperature data includes the temperature inside the cabinet and the temperature of the equipment, reflecting the working environment; the switch status data includes the switch status of each fuse, contactor, etc., reflecting the line connection situation; the protection action data is the action signal of each protection device, reflecting the occurrence of the fault; these data are obtained through the on-site acquisition unit to provide original data support for subsequent fault diagnosis.

[0061] The methods for preprocessing running data include:

[0062] Arrange the operating data of the low-voltage distribution cabinet in chronological order according to the data type (such as voltage data, current data, power data, etc.) to obtain the corresponding data time series, and divide the data time series into several time windows, each time window contains a continuous data point; define a two-dimensional coordinate plane, and for each time window, draw a line graph of the data points on the two-dimensional coordinate plane; add a scan line to the line graph, and place the initial position of the scan line at the first data point; move the scan line step by step from left to right, passing through each data point, and in the process of gradually moving the scan line, record the number of intersections between the horizontal scan line and the line graph.

[0063] Calculate the trend score within the time window corresponding to the data time series based on the number of recorded intersections ;

[0064] ;in, is the change in the number of intersections between the scan line and the line graph under the preset number of moving steps, is the number of data points in the corresponding time window, is the slope of the best fitting line of the line graph, obtained by fitting the line graph using the least squares method. , and is the weight coefficient, is a regulation factor greater than 0; It is a comprehensive determinant that reflects the linearity of the trend.

[0065] Comprehensive determinants ;in, For data points on the line chart The vertical coordinate of For data points on the line chart The horizontal axis of is the intercept of the best-fit straight line of the line graph, is a regulation parameter greater than 0, which makes the comprehensive determination factor present nonlinear jitter. is the slope weight factor, which is adjusted by The value of can control the influence of the slope on the comprehensive determinant; for example, if it is found that some data do not quite meet the linear hypothesis, it can be appropriately increased. The value of , which reduces the comprehensive determinant factor, avoids judging this situation as a good linear fit; on the contrary, if the steepness of the trend is not very sensitive, Setting it to a smaller value makes the slope have less impact on the comprehensive determinant, thereby affecting the value of the entire trend score.

[0066] A trend threshold is preset, and the time window with a trend score greater than the trend threshold is recorded as the trend time window. The number of all trend time windows in the statistical data time series is counted as the trend number. A sequence threshold is preset, and the data type corresponding to the data time series with a trend number greater than the sequence threshold is taken as the data to be processed.

[0067] The data time series corresponding to the data to be processed is used as the root node of the tree to build a segmentation tree; define the cost function , calculate the cost function of the node at the root of the tree The function value of ; denoted as split value;

[0068] ;in, is the residual sum of squares of the corresponding node, is the number of leaf nodes in the current segmentation tree, that is, the number of segments, is the preset maximum number of segments, is the average segment length, and is the baseline parameter, which is adjusted adaptively with the changes in the number and length of segments. For the Segment (leaf node) length, is the penalty coefficient, which can penalize the situation where the data fluctuates too much and make the fitting smoother; is the length penalty coefficient, which can penalize the situation where the segment length difference is too large, making the segmentation more balanced; is the absolute value of the difference between adjacent data points of the corresponding node; the cost function can better balance the fitting quality, model complexity, segmentation balance and data smoothness, and provide a more reasonable evaluation standard for time series segmentation; if the split value is greater than the preset split threshold, the node of the root of the tree is split into two to generate two child nodes, and repeat until the function values ​​of the cost functions of all nodes are less than the split threshold.

[0069] Traversing all the leaf nodes in the segmentation tree, it needs to be explained that the splitting process will eventually generate some terminal nodes, which are leaf nodes; each leaf node corresponds to a segment of the data time series, and the start and end time points of each segment (leaf node) are recorded as the segment boundary. Based on the segment boundary, the data points in the corresponding segment are extracted, and a straight line is fitted using the least squares method. The obtained straight line is the data trend component in the corresponding segment.

[0070] Using the corresponding straight line, for each data point in the corresponding extracted segment, the residual is calculated, and the obtained residuals constitute a residual sequence. The residual sequences of all segments are connected to obtain a stable data sequence corresponding to the data to be processed.

[0071] Perform comprehensive testing on the stationary data sequence; specifically, calculate the stationary data sequence Order autocorrelation coefficient ; Usually 20-40 is used. Too small a value will result in insufficient test power, while too large a value will increase the amount of calculation. ;in, For a stationary data series The value at a time point, is the mean of the stationary data series, is the residual attenuation parameter, The smaller the value, the faster the decay; that is, the latest value of the stationary data sequence is given a greater weight, and the weight of the distant value gradually decays; For a stationary data series Specifically, the stationary data sequence is sorted from early to late, and the data in the first position corresponds to the time point , the second position corresponds to , and so on.

[0072] based on Order autocorrelation coefficient , calculate the preliminary series summary statistics ;

[0073] ;in, is the length of the stationary data sequence (the number of data points); a stationary data sequence is defined as a sequence containing The model with parameters is generated, then the comprehensive statistics of the series are calculated .

[0074] Define the significance level B (usually 0.05 or 0.01), and look up the chi-square distribution table to find the degree of freedom The chi-square distribution critical value ;like Less than , the corresponding stationary data sequence is recorded as the preprocessed data; if Greater than or equal to , then adjust the cost function The parameters within the partition tree are rebuilt, and the stationary data sequence corresponding to the data to be processed is obtained again; until the calculated Less than ; All preprocessed data constitute the preprocessed dataset.

[0075] Methods for extracting features from preprocessed data sets include:

[0076] The preprocessed data set is mapped to a high-dimensional space to obtain multidimensional features corresponding to the time step. Each dimension in the multidimensional features corresponds to a feature, such as voltage, current, power, etc.; all multidimensional features constitute a multidimensional feature set.

[0077] The multidimensional feature set is divided into a training set and a test set in a ratio of 7:3; the training set data is gridded, and the number of grids M2 and the clock period T1 are defined; the multidimensional features in the training set and the test set are divided into a number of clock windows of length T1 in the order of time steps; the multidimensional features in each clock window are discretized, and the values ​​of the multidimensional features are mapped to the indexes of the discrete grids; and the number of times each grid appears in the clock window is counted to obtain a grid count vector.

[0078] Set a grid threshold, which is usually taken as the square root of the number of grids M2. For each clock window, obtain its grid count vector, and traverse each element v_j in the grid count vector, where j is the index of the element; if v_j is greater than or equal to the grid threshold, quantize the element v_j to 1, indicating that the grid appears more times in the clock window; if v_j is less than the grid threshold, quantize the element v_j to 0, indicating that the grid appears less times in the clock window; then obtain the quantized grid count vector, which is the grid pattern of the clock window, and count the grid patterns of all time windows to obtain a grid pattern set.

[0079] Build a clock model and define the number of states , take the size of the grid pattern set, initialize The state transition probability matrix A, in which the element A(u, r) represents the state transition probability from state u to state r, and the initial value is set to ;initialization The emission probability matrix C, where C(r, p) represents the emission probability of emitting to grid p in state r, and the initial value is set to .

[0080] The training set is used to estimate the parameters of the clock model, and the state transition probability and the emission probability are estimated by the expectation maximization (EM) algorithm until a preset maximum number of iterations is reached; thus, a trained clock model is obtained.

[0081] The grid count vectors of the test set in all time windows are connected to form a grid count sequence O. Based on the grid count sequence O, the Viterbi algorithm is used to calculate the most likely state sequence on the trained clock model, and the most likely state sequence is used as the fault feature vector. Each value in the most likely state sequence corresponds to a possible hidden state. The most likely state sequence reflects the state migration pattern of the observed data in time, and is used as a fault feature for subsequent fault identification.

[0082] The methods for optimizing the fault feature vector include:

[0083] Initialize optimization parameters, including population size , Iterations , propagation constant , Water wave height control parameters , Water wave length control parameters .

[0084] Encode the fault feature vector into a binary string and randomly generate binary individuals constitute the initial population; for each binary individual, it is decoded into the corresponding fault feature vector, and the decoded fault feature vector is used as historical data to construct a historical data group, which includes several fault feature vectors and labels corresponding to the fault feature vectors; the label is the status of the low-voltage distribution cabinet (such as minor fault, fault, etc.); the historical data group is divided into a training data set and a verification data set.

[0085] Use a classifier (such as SVM, etc.) to train on the training data set to obtain a classification model, test the classification model on the validation data set, and calculate the classification accuracy as the water wave value corresponding to the binary individual; iteratively update each binary individual. Specifically, sort all corresponding binary individuals from small to large according to the water wave value. For the first binary individuals, calculate their water wave values ; Based on its water wave value Calculate the propagation distance ;

[0086] ;in, is a constant that controls the base size of dynamic adjustment. is the current iteration number, To influence the constant, control the influence of the population fitness difference on the dynamic adjustment; is the maximum difference in water wave values ​​between binary individuals in the population.

[0087] Based on the calculated propagation distance , generating new binary individual solutions ;

[0088] ;in, To represent the first The current solution of binary individuals is a vector (fault feature vector), is the solution of the binary individual with the highest water wave value in the current population (i.e., the global optimal solution, which is a vectors of the same dimension); For one dimensional standard normally distributed random vector, is the dimension of the solution; Perform boundary processing to ensure that the solution is within the legal solution space and calculate The corresponding water wave value new_fit of the binary individual.

[0089] If new_fit is greater than , then update the The solution for binary individuals is , otherwise, calculate the water wave height ;in, is a positive constant that controls the steepness of the tanh function. It is a turning parameter between 0 and 1, which is used to adjust the turning point of the tanh function. The hyperbolic tangent function (tanh) enhances the nonlinear characteristics of the formula and helps the algorithm to escape from the local optimal solution.

[0090] Based on water wave height , generating another new binary individual solution ;in, is a random number between 0 and 1; Perform boundary processing to ensure that it is within the legal solution space; calculate The corresponding binary individual water wave value new_fot, if new_fot is greater than , then update the The solution for binary individuals is , and record the current optimal solution (with the largest water wave value) and its water wave value; repeat the iteration to update each binary individual until the number of iterations is reached , output the binary individual corresponding to the optimal solution, and decode the corresponding binary individual to obtain the optimized fault feature vector.

[0091] Methods for identifying the fault type and fault location include:

[0092] Construct a fault type library and a fault location library; the fault type library contains various fault types that may occur in the low-voltage distribution cabinet, such as short circuit fault, open circuit fault, ground fault, etc.; the fault location library contains the location information of each component in the low-voltage distribution cabinet, recorded as the fault location; such as busbar, switchgear, cable, etc.

[0093] The optimized fault feature vector is decoded to obtain the corresponding fault feature; the fault type of the low-voltage distribution cabinet is initialized to be empty and the fault position of the low-voltage distribution cabinet is initialized to be empty; each fault type in the fault type library is traversed, the similarity between the fault type and the extracted fault feature is calculated, and the fault type with the greatest similarity to the fault feature is selected as the identified fault type.

[0094] Traverse each fault location in the fault location library, calculate the probability of the current fault location having the fault type according to the identified fault type, and select the fault location with the highest probability as the identification result.

[0095] Regarding the process of decoding the optimized fault feature vector, it should be explained that, looking back at the previous process of feature extraction using the clock model, the final fault feature vector obtained is a most likely state sequence, each state corresponds to a possible hidden state, and the optimized fault feature vector is restored to a state sequence according to the same state mapping relationship. Each state of the state sequence actually corresponds to a specific pattern (the distribution pattern of the multidimensional feature in a time window). According to the correspondence between the state and the lattice pattern, the state sequence is translated into a series of lattice patterns; the sequence of these lattice patterns is the corresponding multidimensional feature change pattern over time, that is, the fault feature.

[0096] Regarding the similarity between the calculated fault type and the extracted fault feature, it needs to be explained that the method adopted is the dynamic time warping distance. Specifically, the template features corresponding to the fault types in the fault type library are defined as time series; the dynamic time warping distance between the sequence of grid patterns and the template features corresponding to the fault types is calculated as the time series, which is used as the similarity between the two.

[0097] The method for calculating the probability of the fault type occurring at the current fault location is as follows:

[0098] Based on historical data, an association model between fault location and fault type is constructed. The association model can be represented in the form of a probability matrix or a Bayesian network. For example, the probability matrix P'(j', i') represents the probability of fault type i' occurring at location j'. At the same time, this association model is updated online so that it can be continuously improved with new fault cases. According to the identified fault type i'', for each fault location in the fault location library, the association model is used to find the probability of fault type i'' occurring at each fault location, and the probability is normalized so that the sum is 1. The fault location corresponding to the maximum probability is selected, which is the identified fault location.

[0099] This embodiment realizes real-time remote monitoring of low-voltage distribution cabinet faults, greatly improves the timeliness and work efficiency of fault monitoring, helps to discover potential faults as soon as possible, avoid the expansion of faults, and ensure the reliable operation of the distribution system; secondly, the system adopts a multi-source heterogeneous data fusion analysis method, which can comprehensively acquire and analyze various operating data such as voltage, current, power, temperature, etc. of the distribution cabinet, fully reflect the actual operating status of the distribution cabinet, provide rich data support for fault diagnosis, and greatly improve the accuracy and reliability of fault identification; thirdly, the system can automatically mine the fault features hidden in the data, overcome the defects of traditional methods that rely on manual experience, improve the effectiveness of fault feature extraction and optimization, thereby further improving the accuracy of fault identification; in addition, it can not only identify the fault type, but also locate the specific location of the fault, provide direct guidance for subsequent fault investigation and maintenance, and can greatly shorten the fault handling time and reduce the economic losses caused by the fault.

[0100] Example 2

[0101] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides a remote monitoring method for fault information of a low-voltage power distribution cabinet, including:

[0102] Step 1: Collect the operation data of the low-voltage distribution cabinet, and pre-process the operation data to obtain a pre-processed data set;

[0103] Step 2: Extract features from the preprocessed data set to obtain a fault feature vector;

[0104] Step 3: Optimize the fault feature vector to obtain an optimized fault feature vector;

[0105] Step 4: Based on the optimized fault feature vector, identify the fault type and fault location; send the identified fault type and fault location to the remote monitoring terminal to achieve remote fault monitoring of the low-voltage distribution cabinet.

[0106] Example 3

[0107] 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 remote monitoring method for fault information of a low-voltage distribution cabinet provided above is implemented.

[0108] Since the electronic device introduced in this embodiment is an electronic device used to implement a remote monitoring method for low-voltage distribution cabinet fault information in the embodiment of the present application, based on the remote monitoring method for low-voltage distribution cabinet fault information introduced in the embodiment of the present application, the technical personnel of this field can understand the specific implementation method of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of the present application is not introduced in detail here. As long as the technical personnel of this field implement the electronic device used in the remote monitoring method for low-voltage distribution cabinet fault information in the embodiment of the present application, it belongs to the scope of protection of this application.

[0109] 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.

[0110] 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 remote monitoring system for low-voltage distribution cabinet fault information, characterized in that: include: The data acquisition and processing module is used to collect the operation data of the low-voltage distribution cabinet and pre-process the operation data to obtain a pre-processed data set; A feature extraction module is used to extract features from the preprocessed data set to obtain a fault feature vector; An optimization module, used for optimizing the fault feature vector to obtain an optimized fault feature vector; The fault identification module identifies the fault type and fault location based on the optimized fault feature vector; and sends the identified fault type and fault location to the remote monitoring terminal; Each module is connected via wired and / or wireless means; The method of preprocessing the operation data includes: Arrange the operation data of the low-voltage distribution cabinet in chronological order according to the data type to obtain the corresponding data time series, and divide the data time series into several time windows, each of which contains a continuous data point; define a two-dimensional coordinate plane, and for each time window, draw a line graph of the data points on the two-dimensional coordinate plane; add a scan line to the line graph, and place the initial position of the scan line at the first data point; move the scan line step by step from left to right, passing through each data point, and in the process of gradually moving the scan line, record the number of intersections between the horizontal scan line and the line graph; Calculate the trend score within the time window corresponding to the data time series based on the number of recorded intersections ; ;in, is the change in the number of intersections between the scan line and the line graph under the preset number of moving steps, is the number of data points in the corresponding time window, is the slope of the best-fit straight line of the line graph, , and is the weight coefficient, is a regulation factor greater than 0; is a comprehensive determining factor; Comprehensive determinants ;in, For data points on the line chart The vertical coordinate of For data points on the line chart The horizontal axis of is the intercept of the best-fit straight line of the line graph, is a tuning parameter greater than 0, is the slope weight factor; A trend threshold is preset, and the time window with a trend score greater than the trend threshold is recorded as the trend time window. The number of all trend time windows in the statistical data time series is recorded as the trend number. A sequence threshold is preset, and the data type corresponding to the data time series with a trend number greater than the sequence threshold is taken as the data to be processed; the data time series corresponding to the data to be processed is taken as the node of the tree root to construct a segmentation tree; and a stationary data sequence is obtained; a comprehensive test is performed on the stationary data sequence to obtain the preprocessed data, and all the preprocessed data constitute the preprocessed data set.

2. A remote monitoring system for low-voltage distribution cabinet fault information according to claim 1, characterized in that: The operation data includes voltage data, current data, power data, electric energy data, harmonic data, temperature data, switch status data and protection action data; voltage data includes each phase voltage value and line voltage value; current data includes each phase current value and neutral line current value; power data includes active power, reactive power and apparent power; electric energy data includes active electric energy and reactive electric energy; harmonic data includes the content of each harmonic; temperature data includes cabinet temperature and equipment temperature; switch status data includes the switch status of each fuse and contactor; The protection action data is the action signal of each protection device.

3. A remote monitoring system for low-voltage distribution cabinet fault information according to claim 2, characterized in that: The method of constructing the segmentation tree includes: Define the cost function , calculate the cost function of the node at the root of the tree The function value of ; denoted as split value; ;in, is the residual sum of squares of the corresponding node, is the number of leaf nodes in the current segmentation tree, that is, the number of segments, is the preset maximum number of segments, is the average segment length, and is the benchmark parameter, For the The length of the segment, is the penalty coefficient; is the length penalty coefficient; is the absolute value of the difference between adjacent data points of the corresponding node; If the split value is greater than the preset split threshold, the root node is split into two to generate two child nodes, and this process is repeated until the function values ​​of the cost functions of all nodes are less than the split threshold.

4. A remote monitoring system for low-voltage distribution cabinet fault information according to claim 3, characterized in that: The method of obtaining a stationary data sequence includes: Traverse all leaf nodes in the segmentation tree. Each leaf node corresponds to a segment of the data time series. Record the start and end time points of each segment as the segment boundary. Extract the data points in the corresponding segment based on the segment boundary, and fit a straight line using the least squares method. Use the corresponding straight line to calculate the residual for each data point in the corresponding segment extracted. The obtained residuals constitute a residual sequence. Connect the residual sequences of all segments to obtain the stable data sequence corresponding to the data to be processed.

5. A remote monitoring system for low-voltage distribution cabinet fault information according to claim 4, characterized in that: The method of performing comprehensive detection on the stationary data sequence includes: Calculate the stationary data series Order autocorrelation coefficient ; ;in, For a stationary data series The value at a time point, is the mean of the stationary data series, is the residual attenuation parameter; For a stationary data series The value at each time point; based on Order autocorrelation coefficient , calculate the preliminary series summary statistics ; ;in, is the length of the stationary data sequence; a stationary data sequence is defined as a sequence containing The model with parameters is generated, then the comprehensive statistics of the series are calculated ; Define the significance level B, and find the chi-square distribution table to find the degree of freedom The chi-square distribution critical value ; like Less than , the corresponding stationary data sequence is recorded as the preprocessed data; if Greater than or equal to , then adjust the cost function The parameters within the partition tree are rebuilt, and the stationary data sequence corresponding to the data to be processed is obtained again; until the calculated Less than .

6. A remote monitoring system for low-voltage distribution cabinet fault information according to claim 5, characterized in that: The method of extracting features from the preprocessed data set includes: Map the preprocessed data set to a high-dimensional space to obtain multidimensional features corresponding to the time step, where each dimension in the multidimensional features corresponds to a feature; all multidimensional features constitute a multidimensional feature set; Divide the multidimensional feature set into a training set and a test set; perform grid processing on the training set data, define the number of grids M2 and the clock period T1; divide the multidimensional features in the training set and the test set into a number of clock windows of length T1 in the order of time steps; discretize the multidimensional features in each clock window, map the value of the multidimensional feature to the index of the discrete grid, and count the number of times each grid appears in the clock window to obtain a grid count vector; Set a grid threshold, obtain its grid count vector for each clock window, and traverse each element v_j in the grid count vector, where j is the index of the element; if v_j is greater than or equal to the grid threshold, quantize the element v_j to 1; if v_j is less than the grid threshold, quantize the element v_j to 0; then obtain the quantized grid count vector, which is the grid pattern of the clock window, and count the grid patterns of all time windows to obtain a grid pattern set; Build a clock model and define the number of states , take the size of the grid pattern set, initialize The state transition probability matrix A, in which the element A(u, r) represents the state transition probability from state u to state r, and the initial value is set to ;initialization The emission probability matrix C, where C(r, p) represents the emission probability of emitting to grid p in state r, and the initial value is set to ; Use the training set to estimate the parameters of the clock model, and use the expectation maximization algorithm to estimate the state transition probability and the emission probability until the preset maximum number of iterations is reached; and obtain the trained clock model; Connect the grid count vectors of the test set in all time windows to form a grid count sequence O. Based on the grid count sequence O, use the Viterbi algorithm on the trained clock model to calculate the most likely state sequence, and use the most likely state sequence as the fault feature vector.

7. A remote monitoring system for low-voltage distribution cabinet fault information according to claim 6, characterized in that: The method of optimizing the fault feature vector includes: Initialize optimization parameters, including population size , Iterations , propagation constant , Water wave height control parameters , Water wave length control parameters ; Encode the fault feature vector into a binary string and randomly generate binary individuals constitute the initial population; For each binary individual, decode it into a corresponding fault feature vector, and use the decoded fault feature vector as historical data to construct a historical data set, which includes several fault feature vectors and labels corresponding to the fault feature vectors; the label is the state of the low-voltage distribution cabinet; divide the historical data set into a training data set and a verification data set; Use the classifier to train on the training data set to obtain the classification model, test the classification model on the validation data set, and calculate the classification accuracy as the water wave value corresponding to the binary individual; iteratively update each binary individual until the number of iterations is reached , output the binary individual corresponding to the optimal solution, and decode the corresponding binary individual to obtain the optimized fault feature vector.

8. A remote monitoring system for low-voltage distribution cabinet fault information according to claim 7, characterized in that: The method of iteratively updating each binary individual includes: Sort all corresponding binary individuals according to the water wave value from small to large. binary individuals, calculate their water wave values ; Based on its water wave value Calculate the propagation distance ; ;in, is a constant, is the current iteration number, is the influence constant; is the maximum difference in water wave values ​​between binary individuals in the population; Based on the calculated propagation distance , generating new binary individual solutions ; ;in, To represent the first The current solution of binary individuals, is the solution for the binary individual with the highest water wave value in the current population; For one dimensional standard normally distributed random vector, is the dimension of the solution; right Perform boundary processing and calculate The corresponding binary individual water wave value new_fit; if new_fit is greater than , then update the The solution for binary individuals is , otherwise, calculate the water wave height ;in, is a normal number, is a transition parameter between 0 and 1; Based on water wave height , generating another new binary individual solution ;in, is a random number between 0 and 1; Perform boundary processing; calculate The corresponding binary individual water wave value new_fot, if new_fot is greater than , then update the The solution for binary individuals is , and record the current optimal solution and its water wave value.

9. A remote monitoring system for low-voltage distribution cabinet fault information according to claim 8, characterized in that: The identification method of the fault type and the fault location includes: Construct a fault type library and a fault location library; the fault type library contains various fault types that may occur in the low-voltage distribution cabinet; the fault location library contains the location information of each component in the low-voltage distribution cabinet, recorded as the fault location; decode the optimized fault feature vector to obtain the corresponding fault feature; initialize the fault type of the low-voltage distribution cabinet to be empty and initialize the fault location of the low-voltage distribution cabinet to be empty; traverse each fault type in the fault type library, calculate the similarity between the fault type and the extracted fault feature, and select the fault type with the greatest similarity to the fault feature as the identified fault type; Traverse each fault location in the fault location library, calculate the probability of the current fault location having the fault type according to the identified fault type, and select the fault location with the highest probability as the identification result.

Images