A method and system for monitoring fire hazards of low-voltage lines in a distribution network

Abstract

The present invention relates to the field of data analysis technology, and discloses a method and system for monitoring fire hazards of low-voltage lines in distribution networks, which are used to improve the accuracy of fire hazard monitoring of low-voltage lines in distribution networks. The method includes: real-time acquisition and processing of multi-dimensional parameters of multiple monitoring points of low-voltage lines in distribution networks to obtain an original monitoring data set; transmitting the original monitoring data set, and performing edge computing processing through a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream; integrating the pre-processed multi-source heterogeneous data stream to obtain a fused data set; analyzing the fused data set to obtain a feature vector set reflecting the state of the low-voltage line; performing abnormal state analysis on the feature vector set to obtain the abnormal state and potential fault type of the low-voltage line; performing risk analysis and predictive maintenance decision analysis on the abnormal state and the potential fault type to obtain fire hazard warning information and maintenance suggestions.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a method and system for monitoring fire hazards of low-voltage lines in a distribution network. Background Art

[0002] With the rapid development of the power system and the acceleration of urbanization, the scale and complexity of low-voltage lines in distribution networks are increasing. The traditional method of monitoring fire hazards in low-voltage lines in distribution networks mainly relies on manual inspections and regular inspections. This method has problems such as low monitoring frequency, limited coverage, and poor real-time performance. In recent years, with the development of Internet of Things technology and artificial intelligence, some intelligent monitoring systems have begun to be applied to the monitoring of fire hazards in low-voltage lines in distribution networks. These systems collect parameters such as current, voltage, and temperature of the lines in real time by installing various sensors, and use data analysis technology to process and analyze the collected data to identify potential fire hazards.

[0003] However, the existing intelligent monitoring systems still have some shortcomings. First, most systems only focus on the monitoring of a single or a few parameters, which makes it difficult to fully reflect the operating status of the line. Secondly, the data transmission and processing capabilities are limited, making it difficult to cope with large-scale, high-frequency data collection needs. Furthermore, the data analysis methods are relatively simple and cannot effectively process complex multi-source heterogeneous data, resulting in insufficient accuracy and timeliness in the identification of fire hazards. In addition, most existing systems lack predictive maintenance and decision support functions, making it difficult to provide timely and effective maintenance recommendations for operation and maintenance personnel. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a method and system for monitoring fire hazards of low-voltage lines in a distribution network, which are used to improve the accuracy of monitoring fire hazards of low-voltage lines in a distribution network.

[0005] The present invention provides a method for monitoring fire hazards of low-voltage lines in a distribution network, comprising: real-time acquisition and processing of multi-dimensional parameters of multiple monitoring points of the low-voltage lines in the distribution network to obtain an original monitoring data set; transmitting the original monitoring data set through an adaptive mesh network protocol, and performing edge computing processing through a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream; integrating the pre-processed multi-source heterogeneous data stream through a multi-level data fusion algorithm to obtain a fused data set; analyzing the fused data set through a time domain-frequency domain-time-frequency domain multi-dimensional feature extraction algorithm to obtain a feature vector set reflecting the state of the low-voltage line; performing abnormal state analysis on the feature vector set to obtain the abnormal state and potential fault type of the low-voltage line; performing risk analysis and predictive maintenance decision analysis on the abnormal state and the potential fault type to obtain fire hazard warning information and maintenance suggestions.

[0006] The present invention also provides a fire hazard monitoring system for a distribution network low-voltage line, comprising:

[0007] The acquisition module is used to collect and process multi-dimensional parameters of multiple monitoring points of the distribution network low-voltage line in real time to obtain the original monitoring data set;

[0008] A transmission module, used to transmit the original monitoring data set through an adaptive mesh network protocol, and perform edge computing processing through a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream;

[0009] The integration module is used to integrate the pre-processed multi-source heterogeneous data streams through a multi-level data fusion algorithm to obtain a fused data set;

[0010] An analysis module is used to analyze the fused data set through a time domain-frequency domain-time-frequency domain multi-dimensional feature extraction algorithm to obtain a feature vector set reflecting the state of the low-voltage line;

[0011] An identification module, used for performing abnormal state analysis on the feature vector set to obtain the abnormal state and potential fault type of the low-voltage line;

[0012] The decision-making module is used to perform risk analysis and predictive maintenance decision analysis on the abnormal state and the potential fault type, and obtain fire hazard warning information and maintenance suggestions.

[0013] In the technical solution provided by the present invention, multi-dimensional parameters of multiple monitoring points of the distribution network low-voltage line are collected and processed in real time to obtain the original monitoring data set, thereby realizing comprehensive and real-time monitoring of the line status and greatly improving the comprehensiveness and timeliness of data collection; an adaptive mesh network protocol is used for data transmission, and edge computing processing is performed in combination with a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream. This method significantly improves the efficiency and reliability of data transmission, and at the same time, the preliminary processing and screening of data are realized through edge computing, which effectively reduces the burden on the central processing system and improves the response speed and processing capacity of the overall system; thirdly, the pre-processed multi-source heterogeneous data stream is integrated through a multi-level data fusion algorithm. The fused data set is obtained, which makes full use of the complementarity and correlation between different data sources, improves the quality and reliability of the data, and lays a solid foundation for subsequent analysis; Fourth, the time domain-frequency domain-time-frequency domain multi-dimensional feature extraction algorithm is used to analyze the fused data set to obtain a feature vector set reflecting the state of the low-voltage line. This multi-dimensional feature extraction method can comprehensively capture various characteristics of the line state and improve the accuracy and comprehensiveness of state recognition; Fifth, the abnormal state analysis of the feature vector set is performed to obtain the abnormal state and potential fault type of the low-voltage line. Using advanced data analysis technology, accurate identification of abnormal line states and accurate judgment of fault types are achieved, which greatly improves the accuracy and timeliness of fire hazard identification.

[0014] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 It is a flow chart of a method for monitoring fire hazards of a distribution network low-voltage line in an embodiment of the present invention;

[0017] Figure 2 The figure is a schematic diagram of a fire hazard monitoring system for a low-voltage line of a distribution network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present invention. The following embodiments are used to illustrate the present invention, but cannot be used to limit the scope of the present invention.

[0019] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0020] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0021] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , Figure 1 FIG. 1 is a flow chart of a method for monitoring fire hazards of a distribution network low-voltage line according to an embodiment of the present invention. Figure 1 As shown, the following steps are included:

[0022] S101, collecting and processing multi-dimensional parameters of multiple monitoring points of the distribution network low-voltage line in real time to obtain an original monitoring data set;

[0023] S102, transmitting the original monitoring data set through an adaptive mesh network protocol, and performing edge computing processing through a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream;

[0024] S103, integrating the preprocessed multi-source heterogeneous data streams through a multi-level data fusion algorithm to obtain a fused data set;

[0025] S104, analyzing the fused data set by using a time domain-frequency domain-time-frequency domain multi-dimensional feature extraction algorithm to obtain a feature vector set reflecting the state of the low-voltage line;

[0026] S105, performing abnormal state analysis on the feature vector set to obtain the abnormal state and potential fault type of the low-voltage line;

[0027] S106. Perform risk analysis and predictive maintenance decision analysis on abnormal conditions and potential failure types to obtain fire hazard warning information and maintenance recommendations.

[0028] It should be noted that by installing sensors at multiple monitoring points, multi-dimensional parameter data such as current, voltage, temperature, partial discharge and vibration are collected in real time. These data are timestamped and aligned to form a synchronized raw data stream. Subsequently, the data is cleaned and detected for outliers, preliminary features are extracted, and harmonic analysis, hotspot identification, pulse feature extraction and spectrum analysis are performed to finally obtain comprehensive fire hazard indicators and form the original monitoring data set. Next, an adaptive mesh network protocol is used to transmit data, which can dynamically adjust routing and bandwidth allocation according to the network topology. Local preprocessing and feature extraction are performed at the edge node to achieve preliminary anomaly detection. Then, data aggregation and in-depth analysis are performed at the regional level to obtain regional-level fire hazard assessment results. After data filtering, compression and time synchronization, a multi-source heterogeneous data stream is formed, which contains key information from each monitoring point.

[0029] The multi-level data fusion algorithm integrates these heterogeneous data. First, data standardization and alignment are performed, and data quality is evaluated and weighted. Then, the underlying features are extracted, and the middle-level feature set is obtained through the multi-source information fusion algorithm. Semantic analysis and association rule mining are then performed to integrate and form a high-level semantic feature set. Finally, the fusion decision result is obtained through abnormal pattern recognition and decision-level fusion algorithm, and uncertainty is quantified and corrected to form the final fusion data set. The time domain-frequency domain-time-frequency domain multi-dimensional feature extraction algorithm conducts in-depth analysis of the fusion data set. Statistical features are extracted in the time domain, harmonic analysis and power spectrum density calculation are performed in the frequency domain, and multi-scale decomposition and energy entropy calculation are performed through wavelet packet transform in the time-frequency domain. After fusion, dimensionality reduction, correlation analysis and feature selection, these features finally form a standardized feature vector set that fully reflects the status of the low-voltage line.

[0030] When analyzing the abnormal state of the feature vector set, the multi-dimensional abnormal judgment threshold is first calculated to perform preliminary abnormal marking. Then, the abnormal pattern cluster is obtained through cluster analysis, and the abnormal feature descriptor is extracted. Based on this, a decision tree is constructed to obtain the preliminary fault classification result. Then, the candidate fault type is determined through fuzzy reasoning and historical case matching. Finally, cause analysis and impact assessment are performed to generate a key fault list and diagnostic report to clarify the abnormal state and potential fault type of the low-voltage line. The last step is risk analysis and predictive maintenance decision-making. Risk level classification and fuzzy comprehensive evaluation are performed on abnormal states and potential fault types to obtain accurate risk levels. Combined with vulnerability analysis and fault tree analysis, key weak links and potential fault paths are determined. The probability distribution of fault occurrence is obtained through probability evaluation, and the fire hazard risk index is comprehensively calculated. Based on this, a hierarchical early warning strategy and emergency response plan are formulated, and a preventive maintenance plan and implementation plan are generated, and finally fire hazard early warning information and maintenance recommendations are formed.

[0031] For example, a monitoring point on a low-voltage line of a distribution network detected abnormal fluctuations in the current of phase A, a significant increase in temperature, and a partial discharge signal. After preprocessing the raw data, the edge computing node found that the current harmonic content exceeded the standard, the temperature rise rate was abnormal, and the discharge pulse frequency increased. This information was transmitted to the regional processing center through the mesh network and fused and analyzed with the data from other monitoring points. Time domain analysis found that the current mean increased by 20% and the standard deviation increased by 50%; frequency domain analysis showed that the third harmonic content reached 15%; time-frequency analysis showed that the energy entropy of the discharge signal increased by 30%. The feature extraction algorithm generates a feature vector reflecting the abnormal state. The abnormal state analysis module compares these features with historical data and determines it as a high-probability event of joint overheating failure. The risk analysis results show that if not handled in time, the probability of fire within 48 hours exceeds 80%.

[0032] By executing the above steps, multi-dimensional parameters of multiple monitoring points of the distribution network low-voltage line are collected and processed in real time to obtain the original monitoring data set, which realizes comprehensive and real-time monitoring of the line status and greatly improves the comprehensiveness and timeliness of data collection; an adaptive mesh network protocol is used for data transmission, and edge computing processing is performed in combination with a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream. This method significantly improves the efficiency and reliability of data transmission. At the same time, preliminary processing and screening of data are realized through edge computing, which effectively reduces the burden on the central processing system and improves the response speed and processing capability of the overall system; thirdly, the pre-processed multi-source heterogeneous data stream is integrated through a multi-level data fusion algorithm to obtain The fused data set is fully utilized to make full use of the complementarity and correlation between different data sources, improve the quality and reliability of the data, and lay a solid foundation for subsequent analysis; Fourth, the time domain-frequency domain-time-frequency domain multi-dimensional feature extraction algorithm is used to analyze the fused data set to obtain a feature vector set reflecting the state of the low-voltage line. This multi-dimensional feature extraction method can comprehensively capture various characteristics of the line state and improve the accuracy and comprehensiveness of state recognition; Fifth, the abnormal state analysis of the feature vector set is performed to obtain the abnormal state and potential fault type of the low-voltage line. The advanced data analysis technology is used to realize the accurate identification of the abnormal state of the line and the accurate judgment of the fault type, which greatly improves the accuracy and timeliness of fire hazard identification.

[0033] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0034] (1) Based on multiple monitoring points, the current, voltage, temperature, partial discharge and vibration data of the distribution network low-voltage line are sampled at multiple points to obtain multi-dimensional raw monitoring data; and the multi-dimensional raw monitoring data are time-stamped and data aligned to obtain a synchronized raw data stream;

[0035] (2) Perform data cleaning and outlier detection on the synchronized raw data stream to obtain a preprocessed data set; and perform data segmentation and feature extraction on the preprocessed data set to obtain preliminary feature data;

[0036] (3) Performing harmonic analysis on the current data in the preliminary feature data to obtain current harmonic features; and performing hot spot identification on the temperature data in the preliminary feature data to obtain temperature anomaly features;

[0037] (4) Extract pulse features from the partial discharge data in the preliminary feature data to obtain discharge mode features; and perform spectrum analysis on the vibration data in the preliminary feature data to obtain equipment vibration features;

[0038] (5) The current harmonic characteristics, temperature anomaly characteristics, discharge mode characteristics and equipment vibration characteristics are integrated to obtain a comprehensive fire hazard index; and the comprehensive fire hazard index is associated with the original monitoring data to obtain the original monitoring data set.

[0039] Specifically, high-precision sensors are installed at multiple key monitoring points to collect multi-dimensional parameters such as current, voltage, temperature, partial discharge and vibration in real time. The sampling frequency is adjusted according to the characteristics of different parameters. The current and voltage sampling frequencies are usually set to above 10kHz to capture high-frequency harmonic components; the temperature sampling frequency can be set to 1Hz, which is sufficient to reflect the temperature change trend; the partial discharge sampling frequency needs to reach the MHz level to accurately capture the discharge pulse; the vibration sampling frequency is generally around 1kHz, which is used to monitor the mechanical state of the equipment. The collected multi-dimensional raw monitoring data is then timestamped to ensure the time correspondence of different parameter data. During the data alignment process, data with different sampling frequencies are unified to the same time scale. The lowest sampling frequency is usually selected as the benchmark, and high-frequency data is downsampled. The synchronized raw data stream obtained in this process lays the foundation for subsequent analysis.

[0040] Data cleaning is performed on the synchronized raw data stream, including removing obvious outliers and filling missing values. Outlier detection uses the moving window method to calculate the mean and standard deviation in each window, and mark data points that exceed 3 times the standard deviation as abnormal. Missing values ​​are filled by linear interpolation. These processes ensure the continuity and reliability of the data. The preprocessed data set is segmented, usually divided into fixed time windows (such as 10 minutes). Time domain features such as mean, standard deviation, peak factor, etc. are extracted for each data segment to form preliminary feature data. The amount of data is greatly reduced while retaining key information.

[0041] Harmonic analysis is performed on the current data in the preliminary feature data, and the fast Fourier transform (FFT) algorithm is used to calculate the content of each harmonic. Focus on the third, fifth and seventh harmonics. The abnormal increase in the content of these harmonics often indicates equipment failure or unbalanced line load. Hotspot identification of temperature data uses a spatial clustering algorithm, such as DBSCAN (density-based clustering algorithm), to identify areas with abnormally high temperatures. The algorithm parameters are set as: neighborhood radius ε=5℃, minimum number of points MinPts=3. This means that at least 3 measuring points with abnormal temperatures within a range of 5℃ are identified as hotspots.

[0042] The pulse feature extraction of partial discharge data involves parameters such as discharge amplitude, frequency and phase angle. The PRPD (phase-resolved partial discharge) spectrum analysis method is used to visualize the distribution of discharge pulses within the AC voltage cycle, thereby identifying different types of discharge patterns. The spectrum analysis of vibration data also uses the FFT algorithm, focusing on the energy distribution in a specific frequency band. For example, abnormal vibration in the 50-100Hz frequency band may indicate loose equipment, while abnormalities in the 300-500Hz frequency band may indicate bearing failure.

[0043] Finally, the current harmonic features, temperature anomaly features, discharge mode features, and equipment vibration features are fused, and the weighted summation method is used to obtain the comprehensive fire hazard index. The weight allocation is based on expert experience and historical data analysis. Usually, the weights of current harmonics and temperature anomaly features are higher because these two items are most strongly correlated with fire risks. The comprehensive fire hazard index is associated with the original monitoring data to form the final original monitoring data set. This data set contains not only the original measurement values, but also various features and indicators obtained after processing and analysis, providing a comprehensive information basis for subsequent anomaly detection and fault diagnosis.

[0044] For example, a monitoring point on a low-voltage line in a distribution network collected 6,000 current sampling points, 600 temperature sampling points, 10 million partial discharge sampling points, and 6,000 vibration sampling points within 10 minutes. During the data cleaning process, 15 current anomalies were found and corrected by linear interpolation. FFT analysis showed that the third harmonic content reached 8.5%, exceeding the normal level. Temperature cluster analysis found a hot spot area with a maximum temperature 12°C higher than the surrounding area. PRPD analysis showed that concentrated discharge pulses appeared in the positive half-cycle of the voltage, indicating that surface discharge may exist. The vibration spectrum showed an obvious peak at 80Hz, with an amplitude three times that of the fundamental frequency. Combining these characteristics, the calculated fire hazard index is 0.75 (the full score is 1), which is a high-risk level.

[0045] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0046] (1) Classify and prioritize the original monitoring data set to obtain classified labeled data, and design the network topology structure for the classified labeled data to obtain a network topology diagram;

[0047] (2) Perform routing planning and bandwidth allocation on the network topology to obtain a data transmission plan, and perform distributed transmission of classified labeled data through an adaptive mesh network protocol to obtain node-level raw data streams;

[0048] (3) Perform local preprocessing and feature extraction on the node-level raw data stream to obtain edge node feature data, and perform preliminary anomaly detection on the edge node feature data to obtain node-level anomaly indicators;

[0049] (4) Regional aggregation and in-depth analysis of edge node feature data and node-level abnormal indicators are performed to obtain regional-level fire hazard assessment results. Data is filtered and compressed based on the regional-level fire hazard assessment results to obtain a streamlined data stream.

[0050] (5) Time synchronization and data association are performed on the streamlined data stream to obtain an associated data stream, and multi-dimensional feature fusion is performed on the associated data stream to obtain fused feature data;

[0051] (6) Perform global abnormal pattern recognition on the fused feature data to obtain the global fire hazard index, and compare and analyze the global fire hazard index with historical data to obtain the fire risk trend;

[0052] (7) Classify the fire risk trend into early warning levels to obtain graded early warning information, and generate emergency response strategies based on the graded early warning information to obtain multi-source heterogeneous data streams.

[0053] Specifically, data classification and priority marking are first performed. Data classification is based on data type (such as current, voltage, temperature, etc.) and importance, while priority marking takes into account the timeliness and abnormality of the data. For example, data with a sudden increase in temperature will be marked as high priority. Subsequently, the network topology structure is designed for the classified and marked data. Considering factors such as the geographical distribution of monitoring points, data volume, and network resources, the optimal network connection method is designed. The network topology diagram usually adopts a tree or grid structure to balance transmission efficiency and network reliability. In topology design, practical factors such as the distance between nodes, obstacles, and electromagnetic interference must also be considered.

[0054] Based on the network topology, routing planning and bandwidth allocation are performed. Routing planning uses dynamic routing algorithms, such as the AODV (Ad hoc On-demand Distance Vector) algorithm, which can adjust the data transmission path in real time according to the network conditions. Bandwidth allocation is dynamically adjusted according to data priority and network load conditions to ensure that important data can be transmitted in a timely manner. These plans and allocations form a specific data transmission plan. Data transmission uses an adaptive mesh network protocol, such as the IEEE802.11s standard. This protocol allows direct communication between nodes without passing through a central node, greatly improving the flexibility and reliability of the network. Through this distributed transmission method, the original data flows in the network to form a node-level original data stream.

[0055] At each network node, the received raw data stream is locally preprocessed and feature extracted. Preprocessing includes data denoising, standardization and other operations, and feature extraction uses corresponding algorithms for different types of data. For example, fast Fourier transform (FFT) is performed on current data to extract harmonic features, and the rate of change and peak value are calculated for temperature data. These processes obtain edge node feature data. Next, preliminary anomaly detection is performed on the edge node feature data. Lightweight anomaly detection algorithms are used, such as the statistical 3σ rule or the density-based LOF (Local Outlier Factor) algorithm. Obvious anomalies can be quickly identified and node-level anomaly indicators can be generated.

[0056] The edge node feature data and node-level anomaly indicators are then aggregated and deeply analyzed at the regional level. Regional aggregation uses a hierarchical clustering algorithm to group together node data with similar geographical locations and electrical characteristics. Deep analysis uses more complex algorithms, such as support vector machines (SVM) or random forests, to comprehensively consider multiple factors and derive regional-level fire hazard assessment results. Based on the regional-level fire hazard assessment results, data filtering and compression are performed. For areas assessed as low risk, only key statistical information is retained; for high-risk areas, more detailed data is retained. Data compression uses lossless compression algorithms, such as Huffman coding, to reduce transmission volume while ensuring data integrity. This process streamlines the data flow.

[0057] The streamlined data stream may experience time delays during transmission, so time synchronization is required. The Network Time Protocol (NTP) is used to ensure the time consistency of different nodes. Data association associates data from different sources through timestamps and spatial location information to form an associated data stream. Multi-dimensional feature fusion is performed on the associated data stream, and dimensionality reduction techniques such as principal component analysis (PCA) or autoencoders are used to compress high-dimensional features into low-dimensional space while retaining key information.

[0058] Global abnormal pattern recognition is performed on fused feature data, using deep learning models such as long short-term memory networks (LSTM) or graph neural networks (GNN) to capture complex spatiotemporal patterns. The recognition results form a global fire hazard index, reflecting the safety status of the entire distribution network low-voltage line. The global fire hazard index is compared and analyzed with historical data, and time series analysis methods such as the ARIMA model are used to predict the future trend of fire risks. This analysis result provides a basis for the classification of warning levels.

[0059] The early warning level is usually divided into four levels: normal, attention, warning and critical. Each level corresponds to a different fire risk probability threshold, such as the normal level corresponds to a risk probability of 0-20%, and the critical level corresponds to a risk probability of 80-100%. Based on the divided early warning levels, the corresponding emergency response strategy is generated, including inspection frequency adjustment, emergency maintenance arrangements, etc. Ultimately, all the results of all these processing and analysis, including raw data, feature data, abnormal indicators, risk assessment and early warning information, together constitute a multi-source heterogeneous data stream, providing comprehensive and in-depth information support for fire hazard monitoring of low-voltage lines in distribution networks.

[0060] For example, a distribution network low-voltage line has 100 monitoring points, and each point generates 1KB of raw data per minute. After data classification, 20% are marked as high priority. The network topology design adopts a three-layer tree structure, with the bottom nodes connected through a mesh network, the middle nodes as regional aggregation points, and the top nodes as central processing units. The routing adopts the AODV algorithm, and the dynamic allocation of bandwidth ensures that high-priority data reaches the central node within 30 seconds. The FFT analysis of the edge node shows that the third harmonic content of node 3 reaches 12%, which is much higher than the normal value of 5%, triggering the node-level abnormality mark. After regional aggregation, it is found that the anomaly is highly correlated with the temperature anomaly of the surrounding nodes, and the regional-level fire hazard assessment result is "high risk". Data compression reduces the amount of data in the area by 40% while retaining key information. Time synchronization ensures that the time error of all data is within 10ms. Multidimensional feature fusion compresses the original 50-dimensional features to 10 dimensions, retaining 95% of the information. The LSTM model for global anomaly pattern recognition predicts that the probability of a fire in the area within the next 24 hours is 75%, triggering a "warning" level warning. The emergency response strategy recommends conducting an on-site inspection within 6 hours and preparing emergency repair equipment.

[0061] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0062] (1) Perform data type identification and standardization on the preprocessed multi-source heterogeneous data stream to obtain a standardized data set, and perform time synchronization and spatial alignment on the standardized data set to obtain an aligned data set;

[0063] (2) Performing data quality assessment on the aligned dataset to obtain a data credibility index, and weighting the data based on the data credibility index to obtain a weighted dataset;

[0064] (3) Extract low-level features from the weighted data set to obtain the underlying feature set, and perform feature-level fusion on the underlying feature set using a multi-source information fusion algorithm to obtain the mid-level feature set;

[0065] (4) Perform semantic analysis and association rule mining on the middle-level feature set to obtain semantic features and association rules, and integrate the semantic features and association rules with the middle-level feature set to obtain a high-level semantic feature set;

[0066] (5) Perform abnormal pattern recognition on the high-level semantic feature set to obtain an abnormal feature subset, and then conduct a comprehensive evaluation of the abnormal feature subset through a decision-level fusion algorithm to obtain a fusion decision result;

[0067] (6) Quantify the uncertainty of the fusion decision results to obtain a credibility score, and modify the fusion decision results based on the credibility score to obtain a fusion data set.

[0068] Specifically, the fire hazard monitoring method for low-voltage lines in the distribution network first performs data type identification and standardization when processing multi-source heterogeneous data streams. Data type identification uses the support vector machine (SVM) classification algorithm to identify different types of data such as current, voltage, temperature, partial discharge and vibration through training models. Standardization uses the Z-score method, and the formula is as follows:

[0069] ;

[0070] in, is the standardized value, is the original value, is the data mean, is the standard deviation. This process unifies data of different dimensions to the same scale and obtains a standardized data set. Subsequently, the standardized data set is time synchronized and spatially aligned. Time synchronization uses the Network Time Protocol (NTP) to ensure that the data timestamps of different monitoring points are consistent. Spatial alignment uses coordinate transformation to map the data of different monitoring points to a unified spatial reference system. After this processing, an aligned data set is obtained.

[0071] The data quality of the aligned data set is evaluated, and the data credibility is calculated using multi-dimensional indicators. The credibility indicator calculation formula is as follows:

[0072] ;

[0073] in, For data credibility, For completeness, For accuracy, For timeliness, For consistency, , , , is the corresponding weight. Based on this credibility index, the data is weighted to obtain a weighted data set. Low-level feature extraction is performed on the weighted data set, and different methods are used for different types of data. For example, fast Fourier transform (FFT) is performed on current data to extract harmonic features, and statistical features such as mean and variance are calculated for temperature data. These features constitute the underlying feature set.

[0074] The multi-source information fusion algorithm performs feature-level fusion on the underlying feature set, using the Dempster-Shafer evidence theory. The fusion formula is as follows:

[0075] ;

[0076] in, is the reliability after fusion, and is the reliability of the two sources of evidence, is the conflict factor. This method can effectively handle the uncertainty of data from different sources and obtain a mid-level feature set. Semantic analysis and association rule mining are performed on the mid-level feature set. Semantic analysis uses a word embedding model, such as Word2Vec, to map numerical features to a semantic space. Association rule mining uses the Apriori algorithm to discover potential associations between features. The obtained semantic features and association rules are integrated with the mid-level feature set to form a high-level semantic feature set.

[0077] The high-level semantic feature set is subjected to abnormal pattern recognition using the Long Short-Term Memory Network (LSTM) model. LSTM can effectively capture abnormal patterns in time series, and the identified anomalies constitute an abnormal feature subset. Subsequently, the abnormal feature subset is comprehensively evaluated through a decision-level fusion algorithm, such as Bayesian fusion, to obtain a fusion decision result. Finally, the uncertainty of the fusion decision result is quantified, and the confidence interval is calculated using the Monte Carlo simulation method to obtain a credibility score. Based on this score, the fusion decision result is corrected to finally obtain a fusion data set.

[0078] For example, a monitoring point on a low-voltage line of a distribution network collected 3,600 current data points, 3,600 voltage data points, and 360 temperature data points within 1 hour. The SVM classification algorithm successfully classified these data with an accuracy of 99.9%. After Z-score standardization, the mean of the current data becomes 0 and the standard deviation is 1. Time synchronization ensures that the timestamp error of different types of data is less than 10 milliseconds. Data quality assessment shows that the integrity of the current data is 100%, the accuracy is 99.8%, the timeliness is 100%, and the consistency is 99.5%. Substituting into the formula, the credibility index of 0.95 is obtained. Low-level feature extraction performs FFT on the current data and finds that the third harmonic content is 6.5% and the fifth harmonic content is 3.2%. The Dempster-Shafer evidence theory integrates current, voltage, and temperature features to obtain a 10-dimensional middle-level feature vector. Semantic analysis associates concepts such as "current fluctuation" and "temperature anomaly" with numerical features, and association rule mining finds that the increase in current harmonic content is 85% associated with temperature increase.

[0079] The LSTM model analyzes the time series data and finds a pattern in which the current harmonic content and temperature are abnormal at the same time, triggering a high alert. Bayesian fusion considers multiple factors and ultimately determines that the fire risk of the monitoring point is 80%. Monte Carlo simulation is performed 1,000 times, and the 95% confidence interval of the fire risk is obtained as [75%, 85%].

[0080] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0081] (1) Segment the fused data set into time windows to obtain time series data segments, and extract statistical features from the time series data segments to obtain time domain statistical features;

[0082] (2) Perform fast Fourier transform on the time series data segments to obtain spectrum data, and perform harmonic analysis and power spectrum density calculation on the spectrum data to obtain frequency domain features;

[0083] (3) Perform multi-scale decomposition of the time series data segments through wavelet packet transform to obtain wavelet coefficients, and then perform energy entropy calculation and singular value decomposition on the wavelet coefficients to obtain time-frequency domain features;

[0084] (4) Fusing the time-domain statistical features, frequency-domain features, and time-frequency-domain features to obtain a multidimensional feature set, and performing dimensionality reduction processing on the multidimensional feature set through the principal component analysis algorithm to obtain a reduced-dimensional feature set;

[0085] (5) Perform correlation analysis on the reduced-dimensional feature set to obtain a feature correlation matrix, and perform feature selection based on the feature correlation matrix to obtain a candidate feature subset;

[0086] (6) Perform nonlinear transformation and normalization on the candidate feature subset to obtain a standardized feature vector set, and use the standardized feature vector set as the feature vector set reflecting the status of the low-voltage line.

[0087] Specifically, time window segmentation is first performed. Using sliding window technology, the continuous data stream is divided into several overlapping time series data segments, each of which usually contains data from a few minutes to a few hours. Statistical features are extracted from these time series data segments, and statistics such as mean, standard deviation, skewness, and kurtosis are calculated. These statistics constitute time domain statistical features and reflect the distribution characteristics of the data in the time dimension. Next, a fast Fourier transform (FFT) is performed on each time series data segment. FFT converts the time domain signal into a frequency domain representation to obtain spectrum data. Harmonic analysis is performed on the spectrum data to calculate the amplitude and phase of each harmonic, with special attention paid to low-order harmonics such as the third, fifth, and seventh orders, because these harmonics are often related to equipment failures or line abnormalities. At the same time, the power spectral density is calculated to reflect the distribution of signal energy in frequency. These analysis results constitute frequency domain features, which can capture periodic anomalies and changes in frequency components.

[0088] The wavelet packet transform is the next key step, which performs multi-scale decomposition of the time series data fragments. Compared with the traditional wavelet transform, the wavelet packet transform performs high-frequency and low-frequency decomposition at each scale, providing a more refined time-frequency analysis. The decomposed wavelet coefficients are then subjected to energy entropy calculation to quantify the energy distribution of different frequency bands. In addition, the wavelet coefficient matrix is ​​subjected to singular value decomposition to extract the main time-frequency features. The results obtained from these processes constitute the time-frequency domain features, which can simultaneously reflect the changing characteristics of the signal in time and frequency. The time-domain statistical features, frequency-domain features, and time-frequency domain features are combined into a multidimensional feature set through feature fusion methods. The fusion method can be a simple feature concatenation or a more complex weighted combination. Since the feature dimension may be very high, the principal component analysis (PCA) algorithm is then used for dimensionality reduction. PCA projects the original features into a new feature space through linear transformation, retains several principal components with the largest variance, and thus obtains a reduced-dimensional feature set. Not only does it reduce the amount of data, but it also removes redundant information and improves the efficiency of subsequent analysis.

[0089] Perform correlation analysis on the reduced-dimensional feature set, calculate the correlation coefficients between features, and form a feature correlation matrix. Based on this matrix, perform feature selection, remove highly correlated redundant features, and retain features with strong complementarity. This process obtains a candidate feature subset and further optimizes the feature space. Finally, the candidate feature subset is subjected to nonlinear transformation and normalization. Nonlinear transformation can capture the nonlinear relationship between features. Commonly used methods include logarithmic transformation and exponential transformation. Normalization unifies features of different dimensions to the same scale, usually using minimum-maximum normalization or Z-score normalization. These processes obtain a standardized feature vector set as the final feature representation reflecting the status of the low-voltage line.

[0090] For example, a low-voltage line monitoring point in a distribution network collected 864,000 current data points (10Hz sampling rate) within 24 hours. The time window segmentation uses a window length of 10 minutes to obtain 144 time series data segments. Statistical features such as mean and standard deviation are calculated for each segment to obtain time domain statistical features. FFT analysis found that in addition to the 50Hz fundamental frequency, there was an obvious peak at 150Hz (third harmonic), with an amplitude of 7% of the fundamental frequency, far exceeding the normal level (usually less than 3%). The wavelet packet transform uses db4 wavelet and 6-layer decomposition. The calculated energy entropy shows an abnormal increase in high-frequency bands, indicating that intermittent discharge may exist. After feature fusion, a 500-dimensional feature vector is obtained. PCA dimension reduction retains 95% of the variance information and reduces the dimension to 50. Correlation analysis found that the third harmonic content is highly correlated with the high-frequency energy entropy (correlation coefficient 0.85). After comprehensive consideration, 35 features are selected to form the final feature vector.

[0091] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0092] (1) Calculate the anomaly detection threshold for the feature vector set to obtain the multi-dimensional anomaly determination threshold, and compare the feature vector set with the multi-dimensional anomaly determination threshold to obtain preliminary anomaly labeling data;

[0093] (2) Perform cluster analysis on the preliminary abnormal labeled data to obtain abnormal pattern clusters, and perform feature extraction on the abnormal pattern clusters to obtain abnormal feature descriptors;

[0094] (3) Construct a decision tree based on the abnormal feature descriptor to obtain the preliminary fault classification result, and perform confidence assessment on the preliminary fault classification result to obtain the fault type probability distribution;

[0095] (4) Perform fuzzy reasoning on the probability distribution of fault types to obtain fuzzy fault diagnosis results, and then perform similarity matching between the fuzzy fault diagnosis results and historical fault cases to obtain candidate fault types;

[0096] (5) Perform cause analysis and impact assessment on candidate fault types to obtain fault severity scores, and prioritize faults according to the fault severity scores to obtain a list of critical faults;

[0097] (6) Generate a fault diagnosis report based on the key fault list to obtain the abnormal status and potential fault type of the low-voltage line.

[0098] Specifically, the multivariate statistical process control (MSPC) method is adopted to analyze the historical normal operation data, calculate the mean and standard deviation of each feature, and then set multiple standard deviation ranges as the multidimensional anomaly judgment threshold. Usually, 3 times the standard deviation is used as the boundary of anomaly judgment. Subsequently, the current feature vector set is compared with these thresholds. If a feature exceeds its corresponding threshold, it is marked as abnormal in the preliminary anomaly marking data. Cluster analysis is performed on the preliminary anomaly marking data, and the density clustering algorithm DBSCAN (density-based clustering algorithm) is used. DBSCAN can recognize clusters of arbitrary shapes and has good robustness to noise data. The algorithm divides density-connected points into the same cluster by setting the neighborhood radius ε and the minimum number of points MinPts, thereby obtaining an abnormal pattern cluster. Feature extraction is performed on each abnormal pattern cluster, and statistics such as the center point, radius, and density of the cluster are calculated. These statistics constitute an abnormal feature descriptor, which is used to describe the characteristics of each abnormal pattern.

[0099] A decision tree is constructed based on the abnormal feature descriptor, using the C4.5 algorithm. The C4.5 algorithm selects the best split feature through the information gain ratio and can handle continuous values ​​and missing values. Each leaf node of the decision tree corresponds to a fault type, thus obtaining a preliminary fault classification result. The confidence of this classification result is evaluated, and the probability of each fault type being correctly classified is calculated using the cross-validation method to obtain the probability distribution of the fault type. Fuzzy reasoning is performed on the probability distribution of the fault type, using the Mamdani fuzzy reasoning system. The system converts the precise probability value into a fuzzy language description such as "very likely", "possible", "unlikely", etc. through the fuzzy rule base to obtain the fuzzy fault diagnosis result. This fuzzy result is matched with the historical fault case library for similarity, and the fuzzy similarity calculation method is used to find the several historical cases with the highest similarity as candidate fault types. The cause analysis and impact assessment of the candidate fault types are performed using the fault tree analysis (FTA) method. FTA constructs a fault tree and analyzes from the top event (such as fire) to the basic event (such as equipment aging, overload, etc.) to calculate the severity of each fault type. The evaluation considers the probability of failure and the severity of potential impact, and comprehensively obtains the failure severity score. According to this score, the failures are prioritized to obtain a list of critical failures.

[0100] Finally, a fault diagnosis report is generated based on the key fault list. The report includes abnormal status description, potential fault type, severity score, recommended measures, etc., providing operation and maintenance personnel with comprehensive information on low-voltage line abnormal status and potential fault types.

[0101] For example, the feature vector of a low-voltage line monitoring point in a distribution network contains 50 dimensions. The abnormality judgment threshold calculated by the MSPC method shows that it is judged as abnormal when the third harmonic content exceeds 7.5% (3 times the standard deviation of the normal range). The current data shows that the third harmonic content reaches 8.2%, so it is marked as abnormal. DBSCAN cluster analysis (parameter setting: ε=0.5, MinPts=5) classifies this abnormal point into the "harmonic anomaly" cluster. Feature extraction shows that the third harmonic content of the center point of the cluster is 8.5% and the cluster radius is 0.3%. Decision tree analysis preliminarily classifies this pattern as "non-linear load fault" with a confidence level of 85%. The fuzzy inference system converts the 85% probability into a "very likely" language description. Matching with the historical case library, three similar cases are found, namely transformer saturation, rectifier fault and high-power electronic equipment interference. FTA analysis shows that the severity scores of these three faults are 75, 60 and 50 (full score 100) respectively.

[0102] The fault diagnosis report generated eventually pointed out that there was a high probability of a "non-linear load fault" at the monitoring point, and the most likely cause was transformer saturation (severity 75 / 100). It was recommended to immediately check the transformer status and replace or repair it if necessary. At the same time, the report also recommended checking the rectifier equipment and high-power electronic equipment on the line to rule out other possible interference sources.

[0103] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0104] (1) Classify the risk levels of abnormal conditions and potential fault types to obtain preliminary risk assessment results, and perform fuzzy comprehensive evaluation on the preliminary risk assessment results to obtain accurate risk levels;

[0105] (2) Conduct vulnerability analysis on the low-voltage lines of the distribution network based on accurate risk levels to obtain a list of key weak links, and conduct fault tree analysis on the list of key weak links to obtain potential fault paths;

[0106] (3) Probabilistically evaluate potential fault paths to obtain the probability distribution of fault occurrence, and perform comprehensive risk calculation based on the probability distribution of fault occurrence and the precise risk level to obtain the fire hazard risk index;

[0107] (4) Formulate a graded warning strategy based on the fire hazard risk index, obtain fire hazard warning information, generate an emergency response plan based on the fire hazard warning information, and obtain preliminary emergency response suggestions;

[0108] (5) Conduct a feasibility assessment on the preliminary emergency response proposal to obtain an optimized emergency response plan, and formulate a preventive maintenance plan based on the optimized emergency response plan to obtain a maintenance task list;

[0109] (6) Prioritize and allocate resources for the maintenance task list to obtain a maintenance implementation plan, and integrate the maintenance implementation plan with the fire hazard warning information to obtain fire hazard warning information and maintenance recommendations.

[0110] Specifically, the analytic hierarchy process (AHP) is used to assign weights to various abnormal states and potential fault types to obtain preliminary risk assessment results. Subsequently, the preliminary results are refined using the fuzzy comprehensive evaluation method to obtain the precise risk level. The calculation formula for fuzzy comprehensive evaluation is as follows:

[0111] ;

[0112] in, is the final fuzzy judgment result. is the weight vector, is the fuzzy relationship matrix, Represents fuzzy synthesis operation. Based on the precise risk level, the vulnerability analysis of the low-voltage line of the distribution network is carried out. The network vulnerability assessment method, such as minimum cut set analysis, is used to identify the key weak links in the system. Fault tree analysis (FTA) is performed on these weak links to build a fault tree model and reveal the potential fault path. The top event of the fault tree is set to "fire occurrence", and each intermediate event and basic event are connected through logic gates.

[0113] The Monte Carlo simulation method is used to evaluate the probability of potential failure paths. This method simulates the probability of occurrence of each basic event through a large number of random sampling, and then calculates the probability of occurrence of the top event (fire). The probability calculation formula is as follows:

[0114] ;

[0115] in, is the probability of the top event, is the probability of the ith basic event, and n is the total number of basic events. Combined with the probability distribution of fault occurrence and the precise risk level, a comprehensive risk calculation is performed to obtain the fire hazard risk index. The calculation formula is:

[0116] ;

[0117] in, is the risk index, is the probability of failure, The severity of the consequences of the failure (converted from the precise risk level). According to the fire hazard risk index, a graded warning strategy is formulated. The risk index is usually divided into four levels: low risk (0-0.3), medium risk (0.3-0.6), high risk (0.6-0.8), and extremely high risk (0.8-1.0). Different levels of fire hazard warning information are generated accordingly. Based on the warning information, the event tree analysis (ETA) method is used to generate an emergency response plan and obtain preliminary emergency disposal suggestions.

[0118] Conduct a feasibility assessment on the preliminary emergency response proposals, use the Delphi method to collect expert opinions, and optimize the emergency response plan based on actual resource constraints. According to the optimized plan, formulate a preventive maintenance plan and obtain a maintenance task list. The task list includes information such as inspection items, execution cycle, and required resources. Finally, prioritize and allocate resources for the maintenance task list. Use a multi-objective optimization algorithm, such as NSGA-II (Non-dominated Sorting Genetic Algorithm II), and consider multiple goals such as task urgency, resource utilization efficiency, and risk reduction effect to obtain the optimal maintenance execution plan. Integrate this plan with the fire hazard warning information to form the final fire hazard warning information and maintenance recommendations.

[0119] For example, a monitoring point of a distribution network low-voltage line detected that the three-phase imbalance reached 15%, exceeding the normal threshold (10%). The preliminary risk assessment result was "medium risk". Fuzzy comprehensive evaluation took into account factors such as line aging and load conditions, and finally obtained an accurate risk level of 0.65 (high risk). Vulnerability analysis identified that the main weak link of the line was the aging insulation layer. Fault tree analysis showed that aging of the insulation layer could cause short circuits, which in turn caused fires. Monte Carlo simulation was performed 10,000 times, and the probability of fire was 0.02. The comprehensive risk calculation obtained a fire hazard risk index of 0.02 × 0.65 = 0.013, which is a low risk level. Nevertheless, considering that three-phase imbalance may cause equipment overheating, a warning information of the "need attention" level was still generated. The preliminary emergency response suggestions included "on-site inspection within 48 hours" and "balanced three-phase load". After the feasibility assessment, the optimization plan suggested "on-site inspection within 24 hours" and added the "thermal imaging scan" project. The preventive maintenance program includes monthly checks on the three-phase load balance and quarterly checks on the condition of the insulation.

[0120] The maintenance task priority ranking takes into account task urgency (weight 0.5), resource consumption (weight 0.3) and risk reduction effect (weight 0.2). The final maintenance execution plan lists "on-site inspection within 24 hours" and "thermal imaging scanning" as the highest priority, and "three-phase load balancing" as the second highest priority.

[0121] The embodiment of the present invention also provides a fire hazard monitoring system for a low-voltage line of a distribution network, such as Figure 2 As shown, the fire hazard monitoring system for the low-voltage line of the distribution network specifically includes:

[0122] The acquisition module 201 is used to collect and process multi-dimensional parameters of multiple monitoring points of the distribution network low-voltage line in real time to obtain an original monitoring data set;

[0123] A transmission module 202 is used to transmit the original monitoring data set through an adaptive mesh network protocol, and perform edge computing processing through a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream;

[0124] An integration module 203 is used to integrate the pre-processed multi-source heterogeneous data streams through a multi-level data fusion algorithm to obtain a fused data set;

[0125] An analysis module 204 is used to analyze the fused data set by using a time domain-frequency domain-time-frequency domain multi-dimensional feature extraction algorithm to obtain a feature vector set reflecting the state of the low-voltage line;

[0126] The identification module 205 is used to perform abnormal state analysis on the feature vector set to obtain the abnormal state and potential fault type of the low-voltage line;

[0127] The decision module 206 is used to perform risk analysis and predictive maintenance decision analysis on the abnormal state and the potential fault type to obtain fire hazard warning information and maintenance suggestions.

[0128] Through the collaborative work of the above modules, multi-dimensional parameters of multiple monitoring points of the distribution network low-voltage lines are collected and processed in real time to obtain the original monitoring data set, which realizes comprehensive and real-time monitoring of the line status and greatly improves the comprehensiveness and timeliness of data collection; an adaptive mesh network protocol is used for data transmission, and edge computing processing is combined with a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream. This method significantly improves the efficiency and reliability of data transmission. At the same time, the initial processing and screening of data are realized through edge computing, which effectively reduces the burden on the central processing system and improves the response speed and processing capacity of the overall system; thirdly, the pre-processed multi-source heterogeneous data stream is integrated through a multi-level data fusion algorithm. , and obtain a fused data set, which makes full use of the complementarity and correlation between different data sources, improves the quality and reliability of the data, and lays a solid foundation for subsequent analysis; Fourth, the time domain-frequency domain-time-frequency domain multi-dimensional feature extraction algorithm is used to analyze the fused data set to obtain a feature vector set reflecting the state of the low-voltage line. This multi-dimensional feature extraction method can comprehensively capture various characteristics of the line state and improve the accuracy and comprehensiveness of state recognition; Fifth, the abnormal state analysis of the feature vector set is performed to obtain the abnormal state and potential fault type of the low-voltage line. Using advanced data analysis technology, accurate identification of the abnormal state of the line and accurate judgment of the fault type are realized, which greatly improves the accuracy and timeliness of fire hazard identification.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring fire hazards of low-voltage lines in a distribution network, characterized in that: include: Collect and process multi-dimensional parameters of multiple monitoring points of the distribution network low-voltage lines in real time to obtain the original monitoring data set; The original monitoring data set is transmitted through an adaptive mesh network protocol, and edge computing processing is performed through a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream; The pre-processed multi-source heterogeneous data streams are integrated through a multi-level data fusion algorithm to obtain a fused data set; The fused data set is analyzed by a time domain-frequency domain-time-frequency domain multi-dimensional feature extraction algorithm to obtain a feature vector set reflecting the state of the low-voltage line, including: performing time window segmentation on the fused data set to obtain time series data segments, and performing statistical feature extraction on the time series data segments to obtain time domain statistical features; performing fast Fourier transform on the time series data segments to obtain spectrum data, and performing harmonic analysis and power spectrum density calculation on the spectrum data to obtain frequency domain features; performing multi-scale decomposition on the time series data segments by wavelet packet transform to obtain wavelet coefficients, and performing power spectrum analysis on the wavelet coefficients. Energy entropy calculation and singular value decomposition are performed to obtain time-frequency domain features; feature fusion is performed on the time-domain statistical features, frequency-domain features and time-frequency domain features to obtain a multidimensional feature set, and the multidimensional feature set is subjected to dimensionality reduction processing by a principal component analysis algorithm to obtain a reduced-dimensional feature set; correlation analysis is performed on the reduced-dimensional feature set to obtain a feature correlation matrix, and feature selection is performed based on the feature correlation matrix to obtain a candidate feature subset; nonlinear transformation and normalization are performed on the candidate feature subset to obtain a standardized feature vector set, and the standardized feature vector set is used as a feature vector set reflecting the state of the low-voltage line; Performing abnormal state analysis on the feature vector set to obtain abnormal state and potential fault type of the low voltage line; Perform risk analysis and predictive maintenance decision analysis on the abnormal state and the potential failure type to obtain fire hazard warning information and maintenance suggestions.

2. The method for monitoring fire hazards of low-voltage lines in a distribution network according to claim 1, characterized in that: The step of collecting and processing multi-dimensional parameters of multiple monitoring points of the distribution network low-voltage line in real time to obtain an original monitoring data set includes: Based on the multiple monitoring points, multi-point sampling is performed on the current, voltage, temperature, partial discharge and vibration data of the distribution network low-voltage line to obtain multi-dimensional original monitoring data; and timestamp marking and data alignment are performed on the multi-dimensional original monitoring data to obtain a synchronized original data stream; Performing data cleaning and outlier detection on the synchronized raw data stream to obtain a preprocessed data set; and performing data segmentation and feature extraction on the preprocessed data set to obtain preliminary feature data; Performing harmonic analysis on the current data in the preliminary characteristic data to obtain current harmonic characteristics; and performing hot spot identification on the temperature data in the preliminary characteristic data to obtain temperature anomaly characteristics; Performing pulse feature extraction on the partial discharge data in the preliminary feature data to obtain the discharge mode feature; and performing spectrum analysis on the vibration data in the preliminary feature data to obtain the equipment vibration feature; The current harmonic characteristics, temperature anomaly characteristics, discharge mode characteristics and equipment vibration characteristics are feature fused to obtain a comprehensive fire hazard index; and the comprehensive fire hazard index is associated with the original monitoring data to obtain the original monitoring data set.

3. The method for monitoring fire hazards of low-voltage lines in a distribution network according to claim 1, characterized in that: The step of transmitting the original monitoring data set through the adaptive mesh network protocol and performing edge computing processing through a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream includes: Performing data classification and priority marking on the original monitoring data set to obtain classified marked data, and performing network topology structure design on the classified marked data to obtain a network topology diagram; Performing routing planning and bandwidth allocation on the network topology map to obtain a data transmission plan, and performing distributed transmission on the classified labeled data through an adaptive mesh network protocol to obtain a node-level original data stream; Performing local preprocessing and feature extraction on the node-level raw data stream to obtain edge node feature data, and performing preliminary anomaly detection on the edge node feature data to obtain node-level anomaly indicators; Performing regional aggregation and in-depth analysis on the edge node feature data and node-level abnormal indicators to obtain regional-level fire hazard assessment results, and performing data filtering and compression based on the regional-level fire hazard assessment results to obtain a streamlined data stream; Performing time synchronization and data association on the simplified data stream to obtain an associated data stream, and performing multi-dimensional feature fusion on the associated data stream to obtain fused feature data; Performing global abnormal pattern recognition on the fused feature data to obtain a global fire hazard index, and comparing and analyzing the global fire hazard index with historical data to obtain a fire risk trend; The fire risk trend is divided into warning levels to obtain graded warning information, and an emergency response strategy is generated based on the graded warning information to obtain a multi-source heterogeneous data stream.

4. The method for monitoring fire hazards of low-voltage lines in a distribution network according to claim 1, characterized in that: The step of integrating the preprocessed multi-source heterogeneous data streams through a multi-level data fusion algorithm to obtain a fused data set includes: Performing data type identification and standardization processing on the preprocessed multi-source heterogeneous data stream to obtain a standardized data set, and performing time synchronization and spatial alignment on the standardized data set to obtain an aligned data set; Performing data quality assessment on the aligned data set to obtain a data credibility index, and weighting the data based on the data credibility index to obtain a weighted data set; Performing low-level feature extraction on the weighted data set to obtain an underlying feature set, and performing feature-level fusion on the underlying feature set through a multi-source information fusion algorithm to obtain a mid-level feature set; Performing semantic analysis and association rule mining on the middle-level feature set to obtain semantic features and association rules, and integrating the semantic features and association rules with the middle-level feature set to obtain a high-level semantic feature set; Performing abnormal pattern recognition on the high-level semantic feature set to obtain an abnormal feature subset, and performing comprehensive evaluation on the abnormal feature subset through a decision-level fusion algorithm to obtain a fusion decision result; The uncertainty of the fusion decision result is quantified to obtain a credibility score, and the fusion decision result is corrected based on the credibility score to obtain a fusion data set.

5. The method for monitoring fire hazards of distribution network low-voltage lines according to claim 1, characterized in that: The step of performing abnormal state analysis on the feature vector set to obtain the abnormal state and potential fault type of the low-voltage line includes: Calculating an anomaly detection threshold for the feature vector set to obtain a multi-dimensional anomaly determination threshold, and comparing the feature vector set with the multi-dimensional anomaly determination threshold to obtain preliminary anomaly marking data; Performing cluster analysis on the preliminary abnormality labeling data to obtain abnormality pattern clusters, and performing feature extraction on the abnormality pattern clusters to obtain abnormality feature descriptors; Building a decision tree based on the abnormal feature descriptor to obtain a preliminary fault classification result, and performing a confidence evaluation on the preliminary fault classification result to obtain a fault type probability distribution; Performing fuzzy reasoning on the probability distribution of the fault type to obtain a fuzzy fault diagnosis result, and performing similarity matching between the fuzzy fault diagnosis result and historical fault cases to obtain a candidate fault type; Performing cause analysis and impact assessment on the candidate fault types to obtain fault severity scores, and prioritizing the faults according to the fault severity scores to obtain a critical fault list; A fault diagnosis report is generated based on the critical fault list to obtain the abnormal state and potential fault type of the low-voltage line.

6. The method for monitoring fire hazards of low-voltage lines in a distribution network according to claim 5, characterized in that: The risk analysis and predictive maintenance decision analysis of the abnormal state and the potential fault type are performed to obtain fire hazard warning information and maintenance recommendation steps, including: The abnormal state and the potential fault type are classified into risk levels to obtain a preliminary risk assessment result, and the preliminary risk assessment result is subjected to fuzzy comprehensive evaluation to obtain an accurate risk level; Based on the precise risk level, a vulnerability analysis is performed on the low-voltage lines of the distribution network to obtain a list of key weak links, and a fault tree analysis is performed on the list of key weak links to obtain potential fault paths; Probabilistically evaluating the potential fault path to obtain a fault occurrence probability distribution, and performing a comprehensive risk calculation in combination with the fault occurrence probability distribution and the precise risk level to obtain a fire hazard risk index; Formulate a graded warning strategy according to the fire hazard risk index, obtain fire hazard warning information, generate an emergency response plan based on the fire hazard warning information, and obtain preliminary emergency response suggestions; Conduct a feasibility assessment on the preliminary emergency response proposal to obtain an optimized emergency response plan, and formulate a preventive maintenance plan based on the optimized emergency response plan to obtain a maintenance task list; The maintenance task list is prioritized and resources are allocated to obtain a maintenance execution plan, and the maintenance execution plan is integrated with the fire hazard warning information to obtain fire hazard warning information and maintenance suggestions.

7. A fire hazard monitoring system for a distribution network low-voltage line, used to execute the fire hazard monitoring method for a distribution network low-voltage line according to any one of claims 1 to 6, characterized in that: include: The acquisition module is used to collect and process multi-dimensional parameters of multiple monitoring points of the distribution network low-voltage line in real time to obtain the original monitoring data set; A transmission module, used to transmit the original monitoring data set through an adaptive mesh network protocol, and perform edge computing processing through a multi-layer edge computing algorithm to obtain a multi-source heterogeneous data stream; The integration module is used to integrate the pre-processed multi-source heterogeneous data streams through a multi-level data fusion algorithm to obtain a fused data set; The analysis module is used to analyze the fused data set through a time domain-frequency domain-time-frequency domain multi-dimensional feature extraction algorithm to obtain a feature vector set reflecting the state of the low-voltage line, including: performing time window segmentation on the fused data set to obtain time series data segments, and performing statistical feature extraction on the time series data segments to obtain time domain statistical features; performing fast Fourier transform on the time series data segments to obtain spectrum data, and performing harmonic analysis and power spectrum density calculation on the spectrum data to obtain frequency domain features; performing multi-scale decomposition on the time series data segments through wavelet packet transform to obtain wavelet coefficients, and performing power spectrum density calculation on the wavelet coefficients. energy entropy calculation and singular value decomposition are performed on the number to obtain time-frequency domain features; feature fusion is performed on the time-domain statistical features, frequency-domain features and time-frequency domain features to obtain a multidimensional feature set, and the multidimensional feature set is subjected to dimensionality reduction processing by a principal component analysis algorithm to obtain a reduced-dimensional feature set; correlation analysis is performed on the reduced-dimensional feature set to obtain a feature correlation matrix, and feature selection is performed based on the feature correlation matrix to obtain a candidate feature subset; nonlinear transformation and normalization are performed on the candidate feature subset to obtain a standardized feature vector set, and the standardized feature vector set is used as a feature vector set reflecting the state of the low-voltage line; An identification module, used for performing abnormal state analysis on the feature vector set to obtain the abnormal state and potential fault type of the low-voltage line; The decision-making module is used to perform risk analysis and predictive maintenance decision analysis on the abnormal state and the potential fault type, and obtain fire hazard warning information and maintenance suggestions.

Images