Insulation fault detection method for fire emergency lighting system

Abstract

The invention discloses an insulation fault detection method for a fire-fighting emergency lighting system, and relates to the technical field of intelligent fault diagnosis, and the method comprises the steps: building a structure model of a three-dimensional power distribution loop, collecting historical insulation detection data and environment parameters, recognizing the insulation state of the power distribution loop, and constructing an insulation state information set; based on the risk loop list, injecting a multi-band detection signal into the power distribution loop, acquiring multi-dimensional electrical response data under different frequencies in real time through an electrical parameter acquisition device arranged at a node of the power distribution loop, and constructing a multi-frequency impedance response feature set; by constructing a three-dimensional power distribution loop structure model, the geometric and electrical topological characteristics of the loop can be fully reflected, and the negligible space coupling influence of a two-dimensional plane model is avoided; by adopting Bayesian dynamic risk assessment and defining a risk level according to historical data, a high-risk loop can be identified and early warned in real time, and the timeliness and reliability of warning are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent fault diagnosis technology, and in particular to a method for detecting insulation faults in a fire emergency lighting system. Background Technology

[0002] With the widespread application of fire emergency lighting systems in building safety, the insulation safety of their power supply and distribution circuits has become a core element in ensuring the long-term stable operation of the system. Currently, the insulation status detection of fire emergency lighting systems mostly adopts traditional periodic manual insulation testing or single-point monitoring methods. These methods can initially identify the risk of insulation degradation, but they have problems such as long detection cycles, inability to reflect dynamic changes, and difficulty in accurately identifying local risks. In recent years, methods combining online monitoring of electrical parameters and data analysis have been proposed. By collecting changes in electrical parameters such as insulation resistance and leakage current, the insulation status can be judged, improving detection efficiency. However, a mature three-dimensional structural modeling and global trend analysis system has not yet been formed.

[0003] In existing technologies, most insulation fault detection methods fail to take into account the complex three-dimensional spatial wiring structure of fire emergency lighting systems and lack effective modeling of the relationship between circuit topology, electrical connections and node spatial distribution. This results in the detection results failing to accurately reflect the globality and correlation of actual insulation degradation. At the same time, most existing methods are based on single sampling or static data analysis, which cannot dynamically track the trend of insulation status evolution over time and make it difficult to achieve early warning of potential hazards. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an insulation fault detection method for fire emergency lighting systems to solve the problems of existing technologies that lack three-dimensional structural modeling and cannot dynamically assess the evolution of insulation status.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an insulation fault detection method for a fire emergency lighting system, which includes: establishing a three-dimensional structural model of a power distribution circuit, collecting historical insulation detection data and environmental parameters, identifying the insulation status of the power distribution circuit, and constructing an insulation status information set; Insulation health feature vectors are extracted based on insulation status information set, and spatial neighbor information propagation and temporal feature fusion are performed through graph neural network model to obtain the insulation status prediction value of each distribution circuit. Based on the predicted insulation status of each distribution circuit, combined with the Bayesian dynamic risk assessment method, risky distribution circuits are dynamically identified and a list of risky circuits is generated. Based on the risk circuit list, multi-frequency detection signals are injected into the power distribution circuit. Multi-dimensional electrical response data at different frequencies are collected in real time by electrical parameter acquisition devices set at the nodes of the power distribution circuit, and a multi-frequency impedance response feature set is constructed. Based on the multi-frequency impedance response feature set, the insulation degradation type of each power distribution circuit is identified, and an insulation fault detection report is generated.

[0007] As a preferred embodiment of the insulation fault detection method for the fire emergency lighting system described in this invention, the steps of establishing a three-dimensional power distribution circuit structural model and collecting historical insulation detection data and environmental parameters are as follows: Collect spatial connection information, electrical connection information, and equipment distribution parameters of each power distribution circuit node to establish a three-dimensional structural model of the power distribution circuit; Based on the structural model of the three-dimensional power distribution circuit, historical insulation test data and environmental parameters of each power distribution circuit were collected at different time periods.

[0008] In a preferred embodiment of the insulation fault detection method for the fire emergency lighting system described in this invention, the steps for identifying the insulation status of the power distribution circuit and constructing an insulation status information set are as follows: Historical insulation test data and environmental parameters are standardized and the insulation status of each power distribution circuit is identified. An insulation status information set is constructed based on the insulation status of the power distribution circuit, historical insulation test data, and environmental parameters.

[0009] As a preferred embodiment of the insulation fault detection method for the fire emergency lighting system described in this invention, the steps of extracting insulation health feature vectors based on the insulation state information set and obtaining the predicted insulation state value of each distribution circuit by performing spatial neighbor information propagation and temporal feature fusion through a graph neural network model are as follows: Based on the insulation state information set, an insulation health feature vector is extracted from the insulation state information set using a feature extraction method. The insulation health feature vector is input into the graph neural network model for graph structure information propagation and feature fusion, and combined with multi-layer graph convolution to output the insulation state prediction value of each distribution circuit at different time series.

[0010] As a preferred embodiment of the insulation fault detection method for the fire emergency lighting system described in this invention, the steps for dynamically identifying risky power distribution circuits and generating a list of risky circuits based on the predicted insulation status values ​​of each power distribution circuit and combined with the Bayesian dynamic risk assessment method are as follows: Based on the predicted insulation status, a Bayesian dynamic risk assessment method is used to obtain the risk score of the power distribution circuit. Based on historical insulation testing data, risk level thresholds are defined. By comparing the risk score of a power distribution circuit with a risk level threshold, power distribution circuits with risk scores higher than the risk level threshold are identified, and a list of risk circuits is generated.

[0011] As a preferred embodiment of the insulation fault detection method for the fire emergency lighting system described in this invention, the method involves: injecting multi-frequency detection signals into the power distribution circuit based on a risk circuit list; real-time acquisition of multi-dimensional electrical response data at different frequencies using electrical parameter acquisition devices installed at the nodes of the power distribution circuit; and constructing a multi-frequency impedance response feature set. The steps are as follows: Based on the risk circuit list, analyze the historical insulation test data and three-dimensional structural model of each power distribution circuit to obtain the insulation test frequency range; Based on the insulation detection frequency range, multiple frequency sequences are set; In each power distribution circuit in the risk circuit list, multi-dimensional electrical response data at different frequencies are collected in real time by electrical parameter acquisition devices set at the nodes of the power distribution circuit. Normalization and spatiotemporal feature extraction are performed on multidimensional electrical response data at different frequencies to generate a multi-frequency impedance response feature set.

[0012] In a preferred embodiment of the insulation fault detection method for the fire emergency lighting system described in this invention, the steps for identifying the insulation degradation type of each power distribution circuit based on the multi-frequency impedance response feature set are as follows: Feature extraction is performed on the multi-frequency impedance response feature set to obtain multi-frequency impedance response feature parameters; Based on the characteristic parameters of multi-frequency impedance response, a typical degradation type criterion library is established, and the characteristic parameters of multi-frequency impedance response of each power distribution circuit are archived in a structured manner. The structured archived multi-frequency impedance response characteristic parameters are matched with a typical degradation type criterion library to identify the insulation degradation type of the power distribution circuit.

[0013] As a preferred embodiment of the insulation fault detection method for the fire emergency lighting system described in this invention, the generation of the insulation fault detection report refers to the generation of an insulation fault detection report by summarizing the insulation degradation type and multi-frequency impedance response feature set of the power distribution circuit through a natural language-based intelligent report generation method.

[0014] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the insulation fault detection method for a fire emergency lighting system as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the insulation fault detection method for a fire emergency lighting system as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by integrating the spatial connection relationship, electrical connection relationship and node equipment distribution into a three-dimensional power distribution circuit structure model, the actual layout of the power distribution circuit is accurately restored, ensuring comprehensive data collection and improving the accuracy and reliability of insulation status identification; by combining the graph neural network model with the Bayesian dynamic risk assessment method, the accuracy and timeliness of dynamic early warning are improved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of an insulation fault detection method for fire emergency lighting systems.

[0019] Figure 2 This is a flowchart for trend analysis and risk identification based on insulation state information sets.

[0020] Figure 3 This is a flowchart for multi-band detection and feature construction based on a risk loop list.

[0021] Figure 4 This is a flowchart for identifying insulation degradation types and generating inspection reports based on feature sets. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for detecting insulation faults in a fire emergency lighting system, comprising the following steps: S1: Establish a three-dimensional structural model of the power distribution circuit, collect historical insulation test data and environmental parameters, identify the insulation status of the power distribution circuit, and construct an insulation status information set.

[0026] Collect spatial connection information, electrical connection information, and equipment distribution parameters of each power distribution circuit node to establish a three-dimensional structural model of the power distribution circuit.

[0027] Furthermore, the collected spatial connection information for each power distribution circuit includes the position coordinates of the power distribution circuit nodes in three-dimensional space, and the topological distribution and arrangement of the connection paths between nodes.

[0028] Collect electrical connection information for each power distribution circuit to obtain the electrical conduction structure, connection direction, and circuit number between all nodes.

[0029] Based on the collected spatial and electrical connection information, the equipment distribution parameters of each power distribution circuit node are collected to clarify the type, purpose, and connection port form of the electrical equipment configured at each node.

[0030] Based on the collected spatial connection information, electrical connection information, and equipment distribution parameters of each power distribution circuit node, a three-dimensional structural model of the power distribution circuit is established by binding parameters and constructing geometric entities using a three-dimensional modeling method.

[0031] Based on the structural model of the three-dimensional power distribution circuit, historical insulation test data and environmental parameters of each power distribution circuit were collected at different time periods.

[0032] Furthermore, the spatial connection information, electrical connection information, and equipment distribution parameters of each distribution circuit node are determined as the basis for distribution circuit topology identification and data acquisition path planning. Based on the spatial and electrical connection information of the distribution circuits, fields of the index mapping table are set, and the spatial and electrical connection information is expanded node by node for each distribution circuit, organized into a two-dimensional data table, forming a distribution circuit index mapping table for data scheduling and acquisition control. Based on the constructed three-dimensional distribution circuit structural model, electrical parameter acquisition devices deployed at the corresponding distribution circuit nodes are invoked at different time periods to collect historical insulation detection data corresponding one-to-one with the distribution circuit topology location.

[0033] Historical insulation test data typically includes parameters that characterize the insulation performance of power distribution circuits, such as insulation resistance, leakage current, and capacitance changes.

[0034] Furthermore, at each time historical insulation test data is collected, environmental parameters consistent with the spatial distribution of the power distribution circuit are collected simultaneously.

[0035] Environmental parameters typically include external factors that affect insulation conditions, such as temperature, humidity, air pressure, and dust concentration.

[0036] Historical insulation test data and environmental parameters are standardized and the insulation status of each power distribution circuit is identified.

[0037] Furthermore, historical insulation test data and environmental parameters are time-aligned and merged according to timestamps to generate a fused data matrix.

[0038] The Z-score standardization method is used to calculate the mean and standard deviation of each column in the fused data matrix. Historical insulation test data and environmental parameters are then converted into a standardized fused data matrix with zero mean and unit variance. Statistical characteristic parameters such as mean, standard deviation, extreme values, and coefficient of variation are extracted from the standardized fused data matrix to construct a set of insulation status discrimination indicators. Based on industry standards and empirical rules, combined with historical insulation test data and historical environmental parameters, risk intervals are set for each indicator in the insulation status discrimination indicator set, forming an insulation status discrimination rule table. The insulation status discrimination indicator set of the distribution circuit to be tested is compared item by item with the risk interval corresponding to each indicator in the insulation status discrimination rule table. If all indicators of the distribution circuit to be tested fall into the same risk level interval, the distribution circuit to be tested is directly determined to be at the corresponding risk level. If the risk levels of the indicators are inconsistent, priority rules are followed to identify the insulation status of the distribution circuit.

[0039] Among them, priority rules, for example, give priority to insulation resistance and leakage current indicators. If the results of insulation resistance and leakage current indicators are given priority, then the one with the higher risk level is taken as the base level.

[0040] Priority will be given to the rate of change of electrical capacity and the coefficient of variation of fluctuations. If both show a trend of rising to a higher risk level, the final risk level will be upgraded by one level.

[0041] If the volatility indicator remains stable over a long period, a risk level reduction of one level is permissible.

[0042] An insulation status information set is constructed based on the insulation status of the power distribution circuit, historical insulation test data, and environmental parameters.

[0043] Furthermore, the insulation status set, historical insulation test data, and environmental parameters of the distribution circuits are time-synchronized to generate multi-source information fusion data with consistent paired tags. The insulation status, historical insulation test data, and environmental parameters of each distribution circuit at the corresponding time point in the multi-source information fusion data are structured to construct an insulation status information set. The distribution circuit number of each distribution circuit is used as the first aggregation dimension, and the timestamp corresponding to each insulation status information set is used as the second aggregation dimension. For each distribution circuit, insulation status information is extracted according to the time sequence to obtain an insulation status information set, which typically includes the circuit number, timestamp, historical insulation test data, environmental parameters, etc. The insulation status information sets are grouped according to the same distribution circuit number. Within each distribution circuit number group, the insulation status information sets are sorted according to the timestamp order. The insulation status information sets aggregated and organized according to the distribution circuit number and timestamp are structured to generate the final insulation status information set.

[0044] S2: Extract insulation health feature vectors based on insulation status information set, and use graph neural network model to perform spatial neighbor information propagation and temporal feature fusion to obtain the insulation status prediction value of each distribution circuit.

[0045] Based on the insulation state information set, insulation health feature vectors are extracted from the insulation state information set using feature extraction methods.

[0046] Furthermore, the historical insulation test data in the insulation status information set are preprocessed. The preprocessing includes outlier detection, missing value handling, and data normalization to ensure that the dimensions of each parameter are consistent and comparable.

[0047] Based on the original parameters such as insulation resistance, leakage current, and insulation aging rate in the insulation status information set, statistical analysis methods are used to extract the mean, variance, maximum value, minimum value, kurtosis, skewness, and periodic fluctuation amplitude features to obtain descriptive statistical characteristics.

[0048] By using time-series analysis, features of time-varying insulation resistance parameters in the insulation state information set are extracted to obtain trend features, seasonal features, periodic change features, and abrupt change features, thereby generating time-series features that reflect the dynamic evolution of insulation state.

[0049] By combining environmental parameters, correlation analysis is used to extract the coupling characteristics between insulation performance and environmental changes, thereby obtaining environmental sensitivity characteristics.

[0050] Principal component analysis was used to compress the redundancy of descriptive statistical features, time series features, and environmental sensitivity features obtained from the insulation state information set, and to screen out the principal feature components that have the greatest impact on insulation health assessment.

[0051] All the main feature components that have the greatest impact on insulation health assessment are concatenated and numbered according to a fixed feature vector structure to generate an insulation health feature vector.

[0052] The insulation health feature vector is input into the graph neural network model for graph structure information propagation and feature fusion, and combined with multi-layer graph convolution to output the insulation state prediction value of each distribution circuit at different time series.

[0053] Furthermore, the training process of the graph neural network model is as follows: The structural model of the three-dimensional power distribution circuit and historical insulation detection data are combined into a data sample set with insulation state labels according to the time series. This set is then input into the graph neural network model. The power distribution circuit nodes are processed through multi-layer graph convolution to perform information propagation and feature fusion with their neighboring nodes, obtaining the power distribution circuit's insulation state labels from the 1st to the 2nd time series. A sequence of predicted insulation state values ​​at each time point. ;; in, Indicates that the power distribution circuit is in the 1st to the 2nd A sequence of predicted insulation state values ​​at each time point. This represents the parameters of the graph convolution output layer. This represents the activation function. Indicates the distribution circuit The set of directly connected neighbor loop nodes, This represents the normalization coefficient, with a value range of [0,1]. This represents the learned weight matrix of the graph convolutional layer. Indicates power distribution circuit In the Time points Insulation health feature vector This represents the bias term of the graph convolution output layer.

[0054] The actual insulation condition is marked as historical marking data, and the distribution circuit is marked from the 1st to the 2nd. The predicted insulation state sequence at each time point is compared with historical labeled data. The loss is calculated using the cross-entropy loss function. Based on the loss calculation results, the gradient of each layer parameter of the graph neural network model is calculated using the backpropagation algorithm. The optimized graph neural network model is then obtained through iterative optimization using the SGD optimizer.

[0055] The insulation health feature vector is input into the optimized graph neural network model to obtain the insulation status prediction value of each distribution circuit at different time series.

[0056] S3: Based on the predicted insulation status of each distribution circuit, combined with the Bayesian dynamic risk assessment method, risky distribution circuits are dynamically identified and a list of risky circuits is generated.

[0057] Based on the predicted insulation status, a Bayesian dynamic risk assessment method is used to obtain the risk score of the power distribution circuit.

[0058] Furthermore, the predicted insulation status of each distribution circuit at different time series is used as the observation input data for the Bayesian dynamic risk assessment method to establish an observation sequence.

[0059] The risk levels of power distribution circuits are set as safe, caution, warning, and danger. The risk level of the power distribution circuit is used as a latent variable in the Bayesian dynamic model. Based on the risk level labeling data of the power distribution circuit under different historical time series, the changes in risk level under any two adjacent time series nodes are statistically analyzed to form a state transition sample set.

[0060] The statistically obtained state transition sample set is normalized into a probabilistic form to form a risk level state transition probability matrix, where each element in the risk level state transition probability matrix represents the state transition probability of the distribution circuit risk level.

[0061] Statistical analysis was performed on historical insulation testing data of power distribution circuits based on four risk levels. The mean and standard deviation of the predicted insulation status values ​​for each risk level category were extracted. A probability model for the observed values ​​corresponding to each risk level category was established, and the observed probabilities of the risk level and observed values ​​of the power distribution circuits were calculated, as follows: ; in, Indicates the first The probability of observation for each risk level. Indicates the first Standard deviation of insulation condition prediction values ​​under each risk level Indicates the first The mean of the predicted insulation condition values ​​under each risk level.

[0062] Using observation probabilities as the core prior and likelihood structure of the Bayesian dynamic model, based on Bayes' theorem and combining the state transition probabilities and observation probabilities of the risk level of each distribution circuit, posterior inferences are performed on the risk level of each distribution circuit at each time point to obtain the posterior probability distribution of each distribution circuit at each time step. Finally, the posterior probability distribution of the distribution circuit's risk level is weighted and its corresponding risk weights are used to calculate the expected value, thus obtaining the risk score of the distribution circuit. The expression is as follows: ; in, This indicates that each distribution circuit is in the [number]th [stage]. Risk score at each point in time, Indicates the distribution circuit in the first Risk level at each point in time, Indicates the first The weights for each risk level are set, with values ​​ranging from [0,1]. Indicates that under the observation conditions it belongs to the first The posterior probability of each risk level. The total number of risk level categories.

[0063] Risk level thresholds are defined based on historical insulation testing data.

[0064] Furthermore, key detection parameters are extracted from historical insulation test data of the power distribution circuit at different time periods to form an initial historical insulation test dataset. These key detection parameters typically include insulation test data that can effectively reflect changes in the insulation status of the power distribution circuit, such as insulation resistance value, leakage current value, and capacitance change value.

[0065] The initial historical insulation test dataset is standardized to obtain a standardized historical insulation test dataset. Based on the standardized historical insulation test dataset, the statistical characteristics of the distribution circuit in each time period are calculated to form a distribution circuit feature index set. Then, the entropy weight method is used to calculate the feature index weights to generate the feature index weight distribution. The distribution circuit feature index set and the feature index weight distribution are weighted to obtain the distribution circuit comprehensive status score sequence. The smoothed distribution circuit comprehensive status score sequence is obtained by removing abnormal fluctuations through sliding window smoothing.

[0066] Furthermore, the K value in the K-means clustering algorithm is set to 4. The smoothed power distribution circuit comprehensive status score sequence is clustered using the K-means clustering algorithm to obtain the cluster label and corresponding cluster center value of each power distribution circuit comprehensive status score. The cluster center values ​​are arranged in descending order to obtain four cluster centers arranged in descending order.

[0067] For adjacent cluster centers, the risk level threshold is defined as the midpoint between the two cluster center values, and the risk score of the entire power distribution circuit is divided into four risk levels: safe, caution, warning, and danger.

[0068] Among them, the safety risk level threshold is The risk level threshold for attention is... ;like ≤ < The risk level threshold for the warning is ;like ≥ The risk level threshold is ; in, The midpoint between the first and second cluster center values. The midpoint between the second and third cluster center values. It is the midpoint between the third and fourth cluster center values.

[0069] By comparing the risk score of a power distribution circuit with a risk level threshold, power distribution circuits with risk scores higher than the risk level threshold are identified, and a list of risk circuits is generated.

[0070] Furthermore, the predicted insulation status values ​​of the distribution circuits at different time series are processed by the Bayesian dynamic risk assessment method to obtain a risk score set for the distribution circuits. The risk score of each distribution circuit is compared with the risk level threshold. If the risk score of the distribution circuit is higher than its corresponding risk level threshold, the distribution circuit number is included in the risk circuit number set. A risk circuit list is generated based on the risk circuit number set.

[0071] S4: Based on the risk circuit list, multi-frequency detection signals are injected into the power distribution circuit. Multi-dimensional electrical response data at different frequencies are collected in real time by electrical parameter acquisition devices set at the nodes of the power distribution circuit, and a multi-frequency impedance response feature set is constructed.

[0072] Based on the risk circuit list, we analyze the historical insulation test data and three-dimensional structural model of each distribution circuit to obtain the insulation test frequency range.

[0073] Furthermore, the historical insulation monitoring data of each distribution circuit is uniformly normalized, and wavelet transform analysis is used to decompose the historical insulation monitoring data of each distribution circuit into multi-band detail signals and approximation signals at each scale to obtain the frequency intervals corresponding to each scale. Based on the obtained frequency intervals corresponding to each scale, the energy value of wavelet decomposition of each frequency band is calculated to obtain the energy distribution of different frequency intervals and form a frequency band-energy comparison table. In the frequency band-energy comparison table, the horizontal axis is the frequency interval and the vertical axis is the energy proportion of that frequency band.

[0074] The cumulative energy percentage of each frequency band is calculated, and the frequency bands with higher cumulative energy contribution rates are selected as the main frequency energy concentration ranges.

[0075] By combining the spatial connection relationship and node distribution information of each power distribution circuit in the structural model of the three-dimensional power distribution circuit, the structural feature consistency of the main frequency energy concentration interval selected by wavelet transform analysis is verified.

[0076] If the main frequency energy concentration range falls entirely within the frequency range of the structural feature, then the main frequency energy concentration range is directly used as the final verification range.

[0077] If the frequency range to which the main frequency energy concentration range belongs partially overlaps with the frequency range to which the structural features belong, then the overlapping part is selected as the final verification range.

[0078] If the main frequency energy concentration intervals are completely non-intersecting, then while considering the rationality of the structure, the main frequency energy concentration intervals should be appropriately extended to cover the frequency interval to which the structural characteristics belong.

[0079] Finally, based on the consistency verification results between the main frequency energy concentration range and structural characteristics, the insulation detection frequency range corresponding to each power distribution circuit is obtained.

[0080] Based on the insulation detection frequency range, multiple frequency sequences are set.

[0081] Furthermore, the insulation detection frequency range corresponding to each power distribution circuit is used as the input basis. Based on the structural model of the three-dimensional power distribution circuit, the spatial connection relationship, node distribution information and equipment configuration characteristics of each power distribution circuit are extracted to construct the structural characteristic parameter set of the power distribution circuit.

[0082] By combining the structural characteristic parameters of power distribution circuits with typical electrical fault manifestations, frequency-sensitive areas under different power distribution circuit structures are analyzed, and frequency response templates for typical electrical fault manifestations are constructed.

[0083] Cross-analysis is performed between the insulation detection frequency range and the frequency response template of typical electrical faults to identify the set of frequency points that are compatible with the structural characteristics of the power distribution circuit. Finally, based on the frequency coverage and fault differentiation of the set of frequency points, a multi-frequency sequence suitable for each power distribution circuit is set.

[0084] In each distribution circuit of the risk circuit list, multi-dimensional electrical response data at different frequencies are collected in real time by electrical parameter acquisition devices set at the nodes of the distribution circuit.

[0085] Multidimensional electrical response data includes response voltage, response current, phase response, impedance magnitude, impedance phase angle, frequency label, and sampling timestamp parameters.

[0086] Normalization and spatiotemporal feature extraction are performed on multidimensional electrical response data at different frequencies to generate a multi-frequency impedance response feature set.

[0087] Furthermore, according to the multi-frequency sequence, multi-dimensional electrical response data of each power distribution circuit are collected at each frequency point and normalized according to the parameter type.

[0088] Based on the sampled time series, the time-related dynamic change features of the normalized multidimensional electrical response data are extracted to obtain the time dynamic change features of the multidimensional electrical response data.

[0089] The time dynamic variation characteristics of the multidimensional electrical response data are normalized. The normalized time dynamic variation characteristics are used as input, and the dimensionality of the normalized time dynamic variation characteristics is reduced by principal component analysis. The principal components that can explain most of the variance are selected and the information is highly integrated into feature vectors. The principal component feature vectors obtained after PCA processing of each distribution circuit are ordered and numbered according to frequency label, circuit number and sampling time, and summarized into a multi-frequency impedance response feature set.

[0090] S5: Based on the multi-frequency impedance response feature set, identify the insulation degradation type of each power distribution circuit and generate an insulation fault detection report.

[0091] Feature extraction is performed on the multi-frequency impedance response feature set to obtain the multi-frequency impedance response feature parameters.

[0092] Furthermore, a multidimensional response matrix containing parameters such as response voltage, response current, phase response, impedance magnitude, impedance phase angle, frequency label, and sampling timestamp at different frequencies is extracted from the multi-frequency impedance response feature set. The data of each item in the multidimensional response matrix are normalized to a uniform scale to generate a normalized response matrix. The covariance of the normalized response matrix is ​​calculated by principal component analysis to obtain the covariance matrix of the normalized response matrix.

[0093] Perform eigenvalue decomposition and sorting on the covariance matrix to obtain the real eigenvalues ​​arranged in order of size and the corresponding unit eigenvectors.

[0094] Based on each real eigenvalue and its corresponding unit eigenvector, the variance contribution rate of each principal component in the multi-frequency impedance response feature set is calculated, the cumulative contribution rate of each principal component in the multi-frequency impedance response feature set is calculated, and the number of principal components is determined according to the principle that the cumulative contribution rate is not less than 90%, the final multi-frequency impedance response feature parameters are obtained, and a multi-frequency impedance response feature parameter set is constructed.

[0095] Based on the characteristic parameters of multi-frequency impedance response, a typical degradation type criterion library is established, and the characteristic parameters of multi-frequency impedance response of each power distribution circuit are archived in a structured manner.

[0096] Furthermore, feature combinations including response voltage, response current, phase response, impedance magnitude, and impedance phase angle are selected from the multi-frequency impedance response characteristic parameters as a candidate set of typical degradation characteristics. These characteristics are then labeled with historical fault case data to form a multi-frequency impedance response characteristic label lookup table. Based on samples of different insulation degradation types in the multi-frequency impedance response characteristic label lookup table, cluster analysis and statistical modeling methods are used to extract the statistical characteristic intervals and frequency response patterns of various insulation degradation types. Based on the statistical characteristic intervals and response patterns of various insulation degradation types, a standardized typical degradation type criterion library is constructed. Finally, based on the typical degradation type criterion library, the multi-frequency impedance response characteristic parameters of each distribution circuit are labeled, classified, and organized to complete structured archiving.

[0097] The structured archived multi-frequency impedance response characteristic parameters are matched with a typical degradation type criterion library to identify the insulation degradation type of the power distribution circuit.

[0098] Furthermore, the multi-frequency impedance response characteristic parameters of each distribution circuit are input into the typical degradation type criterion library for parameter range matching analysis to obtain a candidate set of insulation degradation types. Combined with the archived multi-frequency impedance response characteristic parameters in the typical degradation type criterion library, a training sample set and a validation sample set for the support vector machine classification model are constructed. The support vector machine classification model is trained using the training sample set to obtain the optimal classification boundary function.

[0099] The training process of the support vector machine is as follows: extract the archived multi-frequency impedance response feature parameters from the typical degradation type criterion library as training samples, label the training samples according to their degradation type, and form data pairs. Each data pair consists of two parameters: a multi-frequency impedance response feature vector and a corresponding degradation type label.

[0100] All training samples are normalized to a uniform scale, and outliers are checked and removed.

[0101] The training samples are divided into training and validation sets according to a certain ratio. The RBF radial basis function is used as the kernel function of the support vector machine, and the kernel function parameters are adjusted through cross-validation.

[0102] Based on the training set data, a support vector machine (SVM) classification model is constructed. The Lagrange dual problem is solved using the RBF radial basis function kernel function and the adjusted kernel function parameters to obtain the support vectors and their Lagrange multipliers. Based on the support vectors and their Lagrange multipliers, the optimal classification boundary function is obtained. The validation set data is then input into the SVM classification model for validation, resulting in a well-trained SVM classification model.

[0103] Then, the multi-frequency impedance response characteristic parameters of the current power distribution circuit are input into the support vector machine classification model to determine the insulation degradation type of the multi-frequency impedance response characteristic parameters of the current power distribution circuit.

[0104] Based on the insulation degradation type and multi-frequency impedance response feature set of the power distribution circuit, an insulation fault detection report is generated by summarizing the data using a natural language-based intelligent report generation method.

[0105] Furthermore, the insulation degradation type of each distribution circuit is used as the descriptive target input into the intelligent report generation method based on natural language. At the same time, the corresponding multi-frequency impedance response feature set is used as supporting information to fill in the analysis process and technical details into the intelligent report generation method based on natural language. The intelligent report generation method is based on a predefined grammatical structure and text organization rules, combined with the classification labels of insulation degradation types and multi-dimensional feature parameters such as impedance magnitude, impedance phase angle and phase response in the multi-frequency impedance response feature set, to construct the content for report text generation. By combining paragraphs, textual diagnostic results are generated for each distribution circuit. Finally, the natural language content of all distribution circuits is summarized to generate an insulation fault detection report.

[0106] Furthermore, the process of defining the predefined grammatical structure and text organization rules is as follows: based on the requirements for generating the insulation degradation test report, the report content is structurally decomposed to determine the constituent elements of the content. These constituent elements typically include the power distribution circuit number, insulation degradation type, multi-frequency impedance response characteristic parameters, and diagnostic conclusions.

[0107] Based on the content generated from the report text, a basic framework of a preset text template is used, which typically includes a title, data summary, analysis process, diagnostic results, and recommendations.

[0108] Based on the writing standards for Chinese technical documents, we formulate grammatical rules that conform to the logic of natural language. These grammatical rules typically include fixed collocations, causal logic, and parallel relationships, and we define standardized sentence templates. The grammar rules, sentence templates, and organization rules are solidified into a callable generation rule base, which serves as the input basis for natural language generation algorithms.

[0109] This embodiment also provides a computer device applicable to the insulation fault detection method for a fire emergency lighting system, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the insulation fault detection method for a fire emergency lighting system as proposed in the above embodiment.

[0110] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0111] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the insulation fault detection method for a fire emergency lighting system as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0112] In summary, this invention accurately recreates the actual layout of the power distribution circuit by integrating spatial connection relationships, electrical connection relationships, and node equipment distribution into a three-dimensional power distribution circuit structure model, ensuring comprehensive data collection and improving the accuracy and reliability of insulation status identification; and by combining a graph neural network model with a Bayesian dynamic risk assessment method, it improves the accuracy and timeliness of dynamic early warning.

[0113] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method of detecting an insulation fault of a fire emergency lighting system, characterized in that, Includes the following steps: A three-dimensional structural model of the power distribution circuit is established, and historical insulation test data and environmental parameters are collected to identify the insulation status of the power distribution circuit and construct an insulation status information set. Insulation health feature vectors are extracted based on insulation status information set, and spatial neighbor information propagation and temporal feature fusion are performed through graph neural network model to obtain the insulation status prediction value of each distribution circuit. Based on the predicted insulation status of each distribution circuit, combined with the Bayesian dynamic risk assessment method, risky distribution circuits are dynamically identified and a list of risky circuits is generated. Based on the risk circuit list, multi-frequency detection signals are injected into the power distribution circuit. Multi-dimensional electrical response data at different frequencies are collected in real time by electrical parameter acquisition devices set at the nodes of the power distribution circuit, and a multi-frequency impedance response feature set is constructed. Based on the multi-frequency impedance response feature set, the insulation degradation type of each power distribution circuit is identified, and an insulation fault detection report is generated.

2. The method of insulation fault detection of a fire emergency lighting system according to claim 1, characterized in that: The steps for establishing a three-dimensional power distribution circuit structural model and collecting historical insulation test data and environmental parameters are as follows: Collect spatial connection information, electrical connection information, and equipment distribution parameters of each power distribution circuit node to establish a three-dimensional structural model of the power distribution circuit; Based on the structural model of the three-dimensional power distribution circuit, historical insulation test data and environmental parameters of each power distribution circuit were collected at different time periods.

3. The method of insulating fault detection of a fire emergency lighting system according to claim 2, characterized in that: The steps for identifying the insulation status of the power distribution circuit and constructing an insulation status information set are as follows: Historical insulation test data and environmental parameters are standardized and the insulation status of each power distribution circuit is identified. An insulation status information set is constructed based on the insulation status of the power distribution circuit, historical insulation test data, and environmental parameters.

4. The insulation fault detection method for a fire emergency lighting system as described in claim 3, characterized in that: The steps are as follows: Insulation health feature vectors are extracted based on the insulation state information set, and spatial neighbor information propagation and temporal feature fusion are performed using a graph neural network model to obtain the predicted insulation state value for each distribution circuit. Based on the insulation state information set, an insulation health feature vector is extracted from the insulation state information set using a feature extraction method. The insulation health feature vector is input into the graph neural network model for graph structure information propagation and feature fusion, and combined with multi-layer graph convolution to output the insulation state prediction value of each distribution circuit at different time series.

5. The method of insulating fault detection of a fire emergency lighting system according to claim 4, characterized in that: The process involves dynamically identifying risky distribution circuits and generating a list of risky circuits based on the predicted insulation status of each distribution circuit, combined with a Bayesian dynamic risk assessment method. The steps are as follows: Based on the predicted insulation status, a Bayesian dynamic risk assessment method is used to obtain the risk score of the power distribution circuit. Based on historical insulation testing data, risk level thresholds are defined. By comparing the risk score of a power distribution circuit with a risk level threshold, power distribution circuits with risk scores higher than the risk level threshold are identified, and a list of risk circuits is generated.

6. The method of insulating fault detection of a fire emergency lighting system according to claim 5, characterized in that: The process involves injecting multi-frequency detection signals into the power distribution circuits based on a risk circuit list. Multi-dimensional electrical response data at different frequencies are collected in real time using electrical parameter acquisition devices located at the nodes of the power distribution circuits, and a multi-frequency impedance response feature set is constructed. The steps are as follows: Based on the risk circuit list, analyze the historical insulation test data and three-dimensional structural model of each power distribution circuit to obtain the insulation test frequency range; Based on the insulation detection frequency range, multiple frequency sequences are set; In each power distribution circuit in the risk circuit list, multi-dimensional electrical response data at different frequencies are collected in real time by electrical parameter acquisition devices set at the nodes of the power distribution circuit. Normalization and spatiotemporal feature extraction are performed on multidimensional electrical response data at different frequencies to generate a multi-frequency impedance response feature set.

7. The method of insulating fault detection of a fire emergency lighting system according to claim 6, characterized in that: The steps for identifying the insulation degradation type of each power distribution circuit based on the multi-frequency impedance response feature set are as follows: Feature extraction is performed on the multi-frequency impedance response feature set to obtain multi-frequency impedance response feature parameters; Based on the characteristic parameters of multi-frequency impedance response, a typical degradation type criterion library is established, and the characteristic parameters of multi-frequency impedance response of each power distribution circuit are archived in a structured manner. The structured archived multi-frequency impedance response characteristic parameters are matched with a typical degradation type criterion library to identify the insulation degradation type of the power distribution circuit.

8. The method of insulating fault detection of a fire emergency lighting system according to claim 7, characterized in that: The generation of the insulation fault detection report refers to the process of summarizing the insulation degradation type and multi-frequency impedance response feature set of the power distribution circuit using an intelligent report generation method based on natural language to generate the insulation fault detection report. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: When the processor executes the computer program, it implements the steps of the insulation fault detection method for the fire emergency lighting system according to any one of claims 1 to 8.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: When the computer program is executed by the processor, it implements the steps of the insulation fault detection method for the fire emergency lighting system according to any one of claims 1 to 8.

Images