CN117854013A

CN117854013A - Fire monitoring system and method for electrical equipment

Info

Publication number: CN117854013A
Application number: CN202410263154.3A
Authority: CN
Inventors: 马永玲; 罗万鑫; 宋媛媛
Original assignee: Shandong Museum
Current assignee: Shandong Museum
Priority date: 2024-03-08
Filing date: 2024-03-08
Publication date: 2024-04-09

Abstract

The invention relates to the technical field of electrical engineering, in particular to a fire monitoring system and a fire monitoring method for electrical equipment. In the invention, the image quality is improved through a dynamic range imaging and image enhancement technology, so that the characteristics of flame and smoke are more obvious, the real-time monitoring and analyzing capability of the operation state and the environmental change of the electrical equipment is enhanced by adopting a convolutional neural network and a space-time autoregressive model, the faults and risks can be predicted and diagnosed in advance through the comprehensive analysis of the operation state and the environmental change of the electrical equipment, the rapid response is realized, the combination of a dynamic Bayesian network and a graph theory algorithm is realized, the precision of risk assessment and prediction is improved, the risk of fire occurrence is reduced, the monitoring of the state and the environmental change of the electrical equipment is enhanced, and the property and public safety are protected.

Description

Fire monitoring system and method for electrical equipment

Technical Field

The invention relates to the technical field of electrical engineering, in particular to a fire monitoring system and a fire monitoring method for electrical equipment.

Background

The technical field of electrical engineering relates to the development, design, testing and application of electrical systems and devices, including power transmission, distribution, control techniques, and the design and application of electronic devices and systems. The progress in the field is significant in improving the efficiency and safety of industrial automation and household appliances. Including continuous analysis of electrical safety, aimed at reducing the risk of fire by technological innovation, ensuring the safety of property.

Among them, a fire monitoring system for electrical equipment is a system specially used for monitoring and preventing electrical equipment fires, the purpose of which is to detect potential fire risks such as overload, short circuit, arc, etc. in electrical equipment and lines, and cause overheating phenomena of fires, so as to take measures in time to prevent the occurrence of fires. Through the system, electrical equipment can be effectively protected, the occurrence of fire accidents is reduced, and the life and property safety of personnel is protected.

Traditional fire monitoring systems rely on simple smoke detectors or temperature sensors, and lack comprehensive real-time analysis and deep learning capabilities for the environment, so that the traditional fire monitoring systems are insufficient in fire early warning and fault diagnosis. The conventional system does not effectively utilize complex data analysis and image processing technologies, so that the response in fire disaster identification and risk assessment is slow, and the conventional system also shows limitation in node analysis and fault propagation path prediction of a network model, so that comprehensive risk prediction and effective fault prevention strategies cannot be provided. Many deficiencies result in inefficiency in fire monitoring and prevention, increasing potential safety hazards and potential economic losses.

Disclosure of Invention

The invention aims to solve the defects in the prior art and provides a fire monitoring system and a fire monitoring method for electrical equipment.

In order to achieve the above purpose, the present invention adopts the following technical scheme: the fire monitoring system for the electrical equipment comprises a biological characteristic recognition module, a space-time data analysis module, a network theory module, a dynamic Bayesian network module, a graph theory fault propagation analysis module, a time sequence analysis module and a flame monitoring module;

the biological feature recognition module is used for extracting features of images of smoke and flame by adopting a convolutional neural network algorithm based on image data of the surrounding environment of the electrical equipment, analyzing the captured images by combining an image classification technology, recognizing the existence of the smoke and the flame, evaluating potential fire risks in the environment and generating a risk evaluation result;

the space-time data analysis module analyzes the relation of data in space and time by adopting a space-time autoregressive model based on the risk assessment result, processes the data by combining a long-term and short-term memory network, analyzes the running condition of the electrical equipment and generates a fault diagnosis result;

the network theory module simulates the node position of the electrical equipment in the network model by adopting a network science method based on the fault diagnosis result, and identifies the path and potential risk area of fire propagation by dynamically simulating the action of the nodes in the network, so as to generate a fire propagation model;

The dynamic Bayesian network module builds a probability model based on a fire spreading model by adopting a dynamic Bayesian network, analyzes the risk state change in the electrical network, updates risk assessment through probability, and generates a real-time risk state by referring to historical data;

the graph theory fault propagation analysis module analyzes the connection structure of the electrical network based on the real-time risk state by adopting a graph theory analysis algorithm, identifies and predicts a risk area of fault propagation, and generates a fault prediction propagation path;

the time sequence analysis module is used for analyzing the relation of time sequence data by adopting an autoregressive moving average model and a time sequence prediction model based on a fault prediction propagation path, predicting the variation trend of an electrical load and generating a load prediction result;

the flame monitoring module adopts a convolutional neural network algorithm to analyze an image sequence based on a load prediction result, extracts flame and smoke characteristics in a learning image through characteristics, comprises texture and shape recognition, performs scene discrimination, and generates a fire analysis and recognition result.

As a further aspect of the present invention, the risk assessment results include probability score of fire occurrence, fire type identification, location of region where fire occurs, fault diagnosis results include type of electrical equipment fault, time point prediction of fault occurrence, spatial location of fault occurrence, fire propagation model includes path of fire propagation, network node score of fire influence, definition of risk region, real-time risk status includes fire risk assessment value updated based on real-time data, multiple types of risk factors, predicted risk trend, fault prediction propagation path includes electrical network branch of fault influence, speed of fault diffusion, identification of control node, load prediction results include trend prediction using load in future time period of electrical equipment, time interval of load fluctuation, predicted load peak, fire analysis and identification results include flame location, flame size, smoke density.

As a further scheme of the invention, the biological characteristic recognition module comprises an image acquisition sub-module, a characteristic analysis sub-module and a risk assessment sub-module;

the image acquisition sub-module is used for acquiring smoke and flame images based on the surrounding environment of the electrical equipment by adopting a dynamic range imaging and image enhancement technology to generate an image data set;

the characteristic analysis submodule analyzes the characteristics of smoke and flame in an image based on an image data set by adopting a deep convolutional neural network and an image characteristic extraction technology to generate a characteristic analysis result;

and the risk assessment sub-module is used for assessing fire risk levels by adopting a support vector machine and a decision tree classifier based on the feature analysis result, constructing a fire risk assessment model of biological features and generating a risk assessment result.

As a further scheme of the invention, the space-time data analysis module comprises a data acquisition sub-module, a model analysis sub-module and a diagnosis result sub-module;

the data acquisition sub-module is used for collecting historical and real-time data, including electrical equipment operation parameters and environment monitoring data, by adopting a real-time data flow acquisition and data processing technology based on a risk assessment result, so as to generate an electrical equipment operation data set;

The model analysis submodule is used for analyzing the relation of data in space and time based on an electric equipment operation data set by adopting a space-time autoregressive model and a long-term and short-term memory network, analyzing the operation condition and the change trend of the electric equipment and generating a model analysis result;

and the diagnosis result submodule adopts an anomaly detection algorithm to carry out iterative analysis on the data based on the model analysis result, identifies an anomaly mode in operation, diagnoses faults of the electrical equipment and generates a fault diagnosis result.

As a further scheme of the invention, the network theory module comprises a node analysis sub-module, a dynamics simulation sub-module and a path identification sub-module;

the node analysis submodule captures the node position of the electrical equipment in the network by adopting a network analysis method based on the fault diagnosis result, evaluates the function of the electrical equipment in the network and generates a node analysis result;

the dynamics simulation sub-module simulates the action of nodes in the electrical network by adopting a dynamics model based on the node analysis result, analyzes the propagation effect under the fault and fire conditions, and generates a dynamics simulation result;

the path recognition sub-module determines fire, fault diffusion paths and risk areas by adopting a graph theory algorithm based on the dynamic simulation result, draws fault propagation paths and generates a fire propagation model.

As a further scheme of the invention, the dynamic bayesian network module comprises a probability building sub-module, a risk analysis sub-module and a state updating sub-module;

the probability modeling submodule adopts a dynamic Bayesian network to evaluate and update the real-time probability of the network state based on the fire propagation model, and combines historical data to perform risk analysis to generate a probability model;

the risk analysis submodule analyzes the risk states and the change trends of nodes in the electrical network based on the probability model through conditional probability query and structured query language of the Bayesian network, identifies risk areas and risk factors and generates a risk state analysis result;

based on the risk state analysis result, the state updating sub-module updates the risk assessment in real time through a real-time data flow analysis technology and an online learning mechanism of a dynamic Bayesian network, reflects the network state and the risk level, and generates a real-time risk state.

As a further scheme of the invention, the graph theory fault propagation analysis module comprises a structure analysis sub-module, a risk prediction sub-module and a path generation sub-module;

the structure analysis submodule adopts a graph theory analysis method based on the real-time risk state, and comprises network topology stability analysis and node influence evaluation, and captures nodes and weak links in an electrical network to generate a network structure analysis result;

The risk prediction submodule analyzes propagation paths of faults and risks in an electrical network based on network structure analysis results by adopting a graph theory algorithm comprising network flow analysis and propagation dynamics models to generate risk prediction results;

the path generation sub-module analyzes a fault occurrence area by referring to the topological structure of the electric network and real-time risk analysis through a path optimization algorithm based on the risk prediction result to generate a fault prediction propagation path.

As a further scheme of the invention, the time sequence analysis module comprises a data arrangement sub-module, a trend analysis sub-module and a prediction model sub-module;

the data sorting sub-module processes the missing values by adopting a data processing technology including abnormal value detection and interpolation based on the fault prediction propagation path, and performs data standardization by using a Z scoring method to generate a sorted data set;

the trend analysis submodule analyzes periodic changes and trend directions in data based on the sorted data set by adopting seasonal decomposition and trend evaluation technology, predicts load changes in a future time period and generates a trend analysis result;

and the prediction model submodule is used for constructing a composite time sequence prediction model based on the trend analysis result and combining an autoregressive moving average model and a long-term and short-term memory network, predicting the variation trend of the electrical load and generating a load prediction result.

As a further scheme of the invention, the flame monitoring module comprises a dynamic flame analysis sub-module, a smoke behavior evaluation sub-module and a scene discrimination sub-module;

the dynamic flame analysis submodule adopts a convolutional neural network algorithm to extract flame characteristics in the image based on a load prediction result, adopts a multilayer filter to extract edge, texture and color information of the flame, carries out hierarchical learning on the characteristics and generates flame characteristic identification data;

the smoke behavior evaluation sub-module adopts a long-period memory network algorithm to analyze the time sequence of the smoke dynamic behavior based on the flame characteristic identification data, processes the change of the smoke characteristics along with time through a neural network, and generates a smoke behavior evaluation result;

the scene discrimination submodule adopts a random forest classification algorithm to discriminate fire scenes from non-fire scenes based on the smoke behavior evaluation result, and analyzes scene characteristics by constructing a plurality of decision trees to generate fire analysis and recognition results.

A fire monitoring method for an electrical apparatus, which is performed based on the fire monitoring system for an electrical apparatus described above, comprising the steps of:

S1: based on the surrounding environment of the electrical equipment, adopting dynamic range imaging and image enhancement technology to collect images of smoke and flame in the environment, extracting key features in the images, iteratively identifying morphological features of the smoke and the flame, evaluating fire risks, determining risk grades by analyzing the image features, and generating a risk evaluation result;

s2: based on the risk assessment result, collecting and arranging operation parameters and environment monitoring data of the electrical equipment by adopting a real-time data flow acquisition technology and a data processing technology, analyzing the operation condition and environment change trend of the electrical equipment, performing fault diagnosis on the electrical equipment by using an anomaly detection algorithm, identifying an anomaly mode in the data, and generating a fault diagnosis result;

s3: based on the fault diagnosis result, analyzing the node position and function of the electrical equipment in the network by adopting a network analysis method, simulating the function of the nodes in the electrical network, determining the diffusion path and risk area of fire and fault by adopting a graph theory algorithm, and generating a fire spreading model;

s4: based on the fire spreading model, a dynamic Bayesian network is adopted, a probability model is built by combining a particle filtering algorithm, the real-time probability of the network state is estimated and updated, risk analysis is carried out by combining historical data, risk areas and change trends in the network are identified and predicted, and a real-time risk state is generated;

S5: based on the real-time risk state, adopting a graph theory analysis method, including network topology stability analysis and node influence evaluation, identifying key nodes and links in a network, and generating a network structure analysis result;

s6: based on the network structure analysis result, adopting a graph theory algorithm, combining network flow analysis and a propagation dynamics model, and carrying out predictive analysis on propagation paths of faults and risks in an electrical network to generate a fault predictive propagation path;

s7: based on the fault prediction propagation path, analyzing periodic changes in the path by adopting a seasonal decomposition technology, determining the trend direction of the path by using a trend evaluation technology, constructing a composite time sequence prediction model by combining an autoregressive moving average model, predicting the electrical load change trend in a future time period, and generating a load prediction result;

s8: based on the load prediction result, a convolutional neural network algorithm is adopted to process images around the electrical equipment, flame characteristic identification data are analyzed, dynamic behavior of smoke is analyzed, a random forest classification algorithm is adopted to analyze a real-time scene, fire conditions are judged and identified, and fire analysis and identification results are generated.

Compared with the prior art, the invention has the advantages and positive effects that:

in the invention, the accuracy and efficiency of fire monitoring are obviously improved by adopting advanced technologies such as dynamic range imaging, image enhancement technology, convolutional neural network, space-time autoregressive model, dynamic Bayesian network, graph theory analysis algorithm and the like. The dynamic range imaging and image enhancement technology improves the image quality, so that the characteristics of flame and smoke are more obvious, the identification is convenient, the convolutional neural network and the application of the space-time autoregressive model are improved, the real-time monitoring and analyzing capability on the running state and environmental change of the electrical equipment is enhanced, the combination of the dynamic Bayesian network and the graph theory algorithm is improved, the precision of risk assessment and prediction is improved, and the fault prevention and the risk control are more effective.

Drawings

FIG. 1 is a system flow diagram of the present invention;

FIG. 2 is a schematic diagram of a system framework of the present invention;

FIG. 3 is a flow chart of a biometric identification module of the present invention;

FIG. 4 is a flow chart of a spatio-temporal data analysis module of the present invention;

FIG. 5 is a flow chart of a network theory module of the present invention;

FIG. 6 is a flow chart of a dynamic Bayesian network module of the present invention;

FIG. 7 is a flow chart of a graph theory fault propagation analysis module of the present invention;

FIG. 8 is a flow chart of a time series analysis module according to the present invention;

FIG. 9 is a flow chart of a flame monitoring module of the present invention;

FIG. 10 is a schematic diagram of the method steps of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

In the description of the present invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientation or positional relationships shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, in the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

Embodiment one: referring to fig. 1 to 2, the present invention provides the following technical solutions: the fire monitoring system for the electrical equipment comprises a biological characteristic recognition module, a space-time data analysis module, a network theory module, a dynamic Bayesian network module, a graph theory fault propagation analysis module, a time sequence analysis module and a flame monitoring module;

The dynamic Bayesian network module builds a probability model based on a fire spreading model by adopting a dynamic Bayesian network, analyzes the risk state change in the electrical network, updates the risk assessment through probability, and generates a real-time risk state by referring to historical data;

the graph theory fault propagation analysis module analyzes the connection structure of the electrical network by adopting a graph theory analysis algorithm based on the real-time risk state, and identifies and predicts the risk area of fault propagation to generate a fault prediction propagation path;

The risk assessment results comprise probability scores of fire occurrence, fire type identification and region positioning of fire occurrence, the fault diagnosis results comprise types of faults of the electrical equipment, time point predictions of fault occurrence and space positions of fault occurrence, the fire propagation model comprises paths of fire propagation, network node scores of fire influence and definition of risk regions, the real-time risk state comprises fire risk assessment values updated based on real-time data, multiple risk factors and predicted risk trends, the fault prediction propagation paths comprise electrical network branches of fault influence, speed of fault diffusion and identification of control nodes, the load prediction results comprise trend predictions of loads used in future time periods of the electrical equipment, time intervals of load fluctuation and predicted load peaks, and the fire analysis and identification results comprise flame positions, flame sizes and smoke densities.

In the biological feature recognition module, the feature extraction of smoke and flame images is performed by using a Convolutional Neural Network (CNN) through image data acquired from the surrounding environment of the electrical equipment. A CNN model is constructed using a TensorFlow framework, which includes a plurality of convolution layers (Conv 2D) and pooling layers (MaxPooling 2D) to extract image features, the features are classified by a full-join layer (tense), an optimizer selects Adam to optimize the network, and a loss function uses cross entropy (cross entropy) to calculate errors. And after model training is completed, performing model accuracy assessment by using the verification set. The generated fire risk assessment model provides a basis for guiding subsequent risk management and emergency measures.

In the space-time data analysis module, a method of combining a space-time autoregressive model with a long-short-term memory network (LSTM) is adopted to process data. The spatiotemporal autoregressive model is responsible for capturing the correlation of data in time and space, and LSTM deals with long-term dependency problems in time series data. The method comprises the steps of setting an LSTM network structure, receiving output from a space-time autoregressive model by an input layer, performing deep learning by a hidden layer through a plurality of LSTM units, and performing fault diagnosis by an output layer based on the learned characteristics. Through training and verification, the model can accurately identify the abnormal operation of the electrical equipment, and generate a fault diagnosis result for indicating the operation state of the equipment and potential fault problems.

In the network theory module, the function of the electrical equipment in the network is analyzed by adopting a network science method and dynamics simulation. A graph model of the electrical network is defined by a network science method, including mapping of nodes and edges, and kinetic simulation is used to study interactions between nodes and their effects on overall network behavior. Kinetic modeling helps understand how information propagates through the network in the event of an emergency, such as a fire, identifying critical nodes and potentially risk areas. The fire propagation model provides detailed views of fire propagation paths and risk areas and provides basis for emergency response and risk mitigation.

In the dynamic bayesian network module, a probability model is built through a fire propagation model to analyze risk state changes in the electrical network. And processing the time sequence data by using a dynamic Bayesian network, and updating the probability estimation of the network state in real time by combining a particle filtering algorithm through the change rule of the historical data learning risk state by using the model. The refinement operation includes setting a network structure, initializing a probability distribution, and applying bayesian rules to update probability estimates. The generated real-time risk state provides instant risk assessment and assists the decision maker in quick response.

In the graph theory fault propagation analysis module, a graph theory analysis algorithm is used for analyzing the connection structure of the electrical network. Through network topology stability analysis and node influence evaluation, the algorithm identifies propagation paths and risk areas of faults. The refinement operation comprises the steps of constructing a graph model, running a path analysis algorithm and calculating an influence score of the node. The generated fault prediction propagation path provides basis for a subsequent risk mitigation strategy, helping to reduce potential losses.

In the time sequence analysis module, an autoregressive moving average model and a time sequence prediction model are adopted to analyze and predict the change trend of the electrical load. The model captures periodic changes and trend directions in the time series by sorting the data sets, applying seasonal decomposition techniques, and trend assessment techniques. The generated load prediction result helps the manager optimize the resource allocation and improves the energy efficiency of the system.

In the flame monitoring module, a convolutional neural network algorithm is used for analyzing an environment image, and features of flames and smoke in the image are learned, including setting a CNN structure, training a model to identify textures and shapes of the flames and the smoke, and applying a classifier to conduct scene discrimination, so that generated fire analysis and identification results can guide an emergency response team in real time, and rapid and effective fire treatment is ensured.

Referring to fig. 2 and 3, the biometric feature recognition module includes an image acquisition sub-module, a feature analysis sub-module, and a risk assessment sub-module;

the feature analysis submodule analyzes the features of smoke and flame in the image based on the image data set by adopting a deep convolutional neural network and an image feature extraction technology to generate a feature analysis result;

based on the feature analysis result, the risk assessment sub-module adopts a support vector machine and a decision tree classifier to assess fire risk level, constructs a fire risk assessment model of biological features, and generates a risk assessment result.

In the image acquisition sub-module, real-time images near the electrical equipment are collected through a camera of the surrounding environment, and the dynamic range imaging technology and the image enhancement algorithm are used for improving the image quality and improving the discernability of smoke and flame. The image data set is stored in a high dynamic range image format, so that the detail is rich, and the subsequent feature extraction and analysis are facilitated. An image dataset comprising smoke and flame features is generated as a basic input to the fire monitoring system, providing raw data for iterative feature analysis and risk assessment.

In the feature analysis submodule, the collected image data set is processed through a deep convolutional neural network, and key smoke and flame features are extracted. The neural network is constructed using the TensorFlow and Keras libraries, and the network structure includes multiple convolution layers and pooling layers in order to capture key features in the image, such as shape, texture, and color. Network parameters are adjusted through a training process, and model performance is continuously optimized by using the loss function and the optimizer. The generated characteristic analysis results describe the characteristics of smoke and flame in the image in detail, and provide accurate input for risk assessment.

In the risk assessment sub-module, the feature analysis result is processed through a support vector machine and a decision tree classifier, and the fire risk level is assessed. The feature analysis results are used as input data, and the support vector machine and decision tree algorithm train classification models based on the data to determine fire risk levels present in the images. The model can accurately classify images into different risk levels through training and verification. The generated risk assessment results are displayed in a structural format, including probability scores and risk grades of fires, and provide basis for subsequent early warning and countermeasure measures.

Referring to fig. 2 and 4, the spatiotemporal data analysis module includes a data acquisition sub-module, a model analysis sub-module, and a diagnosis result sub-module;

the data acquisition sub-module is used for collecting history and real-time data, including electrical equipment operation parameters and environment monitoring data, by adopting a real-time data flow acquisition and data processing technology based on a risk assessment result, so as to generate an electrical equipment operation data set;

the model analysis submodule analyzes the relation of data in space and time based on an electric equipment operation data set by adopting a space-time autoregressive model and a long-period memory network, analyzes the operation condition and the change trend of the electric equipment, and generates a model analysis result;

the diagnosis result submodule adopts an anomaly detection algorithm to carry out iterative analysis on the data based on the model analysis result, identifies an anomaly mode in operation, diagnoses faults existing in the electrical equipment and generates a fault diagnosis result.

In the data acquisition sub-module, historical and real-time data including electrical equipment operation parameters and environment monitoring data such as temperature, humidity, current, voltage and the like are collected according to risk assessment results through real-time data flow acquisition and data processing technologies. The collected data is subjected to preprocessing steps such as cleaning, normalization and the like, so that the quality and consistency of the data are ensured, and finally an operation data set of the electrical equipment is generated. The result is of great importance for monitoring the health condition and the environmental condition of the electrical equipment, and provides a basis for subsequent model analysis and fault diagnosis.

In the model analysis submodule, the electric equipment operation data set is subjected to deep analysis through a space-time autoregressive model and a long-short-term memory network (LSTM). The spatio-temporal autoregressive model helps understand the dynamic correlation of data in time and space, while LSTM deals specifically with long-term dependency problems in time series data. The two methods are combined for use, and the operation condition and the environment change trend of the electrical equipment can be deeply analyzed. Defining an LSTM network structure, receiving space-time data by an input layer, performing deep learning through a plurality of LSTM units, giving out a prediction and trend analysis result of an operation condition by an output layer, and revealing the operation state and potential risk of electrical equipment by a generated model analysis result to provide a basis for fault prediction.

In the diagnosis result submodule, iterative analysis is carried out on a model analysis result through an abnormality detection algorithm so as to identify an abnormal mode and potential faults in operation. Anomaly detection algorithms, such as isolated forests or self-encoders, are used to analyze operational data of electrical devices to identify data points that differ significantly from historical normal behavior patterns. Including setting thresholds for anomaly detection, training a model to distinguish between normal and abnormal conditions, and then applying the model to a new data set to identify anomalies. Through detailed analysis, a fault diagnosis result is finally generated, not only fault points are identified, but also specific types and positions of faults are provided, and clear maintenance guidelines and preventive measures are provided for maintenance teams.

Referring to fig. 2 and 5, the network theory module includes a node analysis sub-module, a dynamics simulation sub-module, and a path identification sub-module;

In the node analysis submodule, information of nodes and edges is analyzed through a network analysis method. The electric network is represented by using the graphical data, key nodes are identified by adopting network science theory such as centrality analysis and the like, and the roles and importance of the nodes in the whole network are evaluated. The method comprises the steps of calculating indexes such as centrality, proximity centrality and medium centrality of the nodes, and revealing positions and influences of the nodes in a network. Through analysis, the generated node analysis result is used for specifying the network function and importance of the node, and providing basic data for subsequent dynamics simulation.

In the dynamics simulation sub-module, interactions and state changes between nodes are analyzed by a dynamics model. Based on the node analysis results, dynamic equations and network models are applied to simulate node behavior in an electrical network and its propagation effects under fault and fire conditions. And adopting a system dynamics model or a network propagation model to simulate node reaction and information propagation speeds and fault diffusion paths under different conditions. The results of the dynamics simulation help understand how the network responds in the event of a particular fault or fire, generating the dynamics simulation results, depicting the context and dynamics of the underlying fault propagation.

In the path identification sub-module, fault diffusion paths and risk areas in the network are analyzed through a graph theory algorithm. Based on the results of the kinetic simulation, a fault or fire propagation path is determined and mapped using a graph theory algorithm such as a shortest path algorithm or network flow analysis. The connection strength and the propagation probability between nodes are analyzed in detail, and key propagation paths and high risk areas are identified. The generated fire spreading model describes the specific paths of faults and fire spread in the electric network in detail, and provides accurate basis for making countermeasures and relieving strategies.

Referring to fig. 2 and 6, the dynamic bayesian network module includes a probability modeling sub-module, a risk analysis sub-module, and a state update sub-module;

the probability modeling submodule adopts a dynamic Bayesian network to evaluate and update the real-time probability of the network state based on the fire spreading model, and combines historical data to perform risk analysis to generate a probability model;

the risk analysis submodule analyzes the risk state and the change trend of nodes in the electrical network based on the probability model through conditional probability query and structured query language of the Bayesian network, identifies risk areas and risk factors and generates a risk state analysis result;

In the probability modeling sub-module, historical events and real-time monitoring data are analyzed through a dynamic Bayesian network, a particle filtering algorithm is adopted to estimate and update the state of the electric network in real time, and the historical data and real-time input are fused to evaluate the real-time probability of the network state. Initializing particle sets is covered, priori knowledge is represented through random sampling, particle weights are updated according to new observation data, and resampling is conducted to avoid weight degradation, so that probability distribution reflecting the current network state is formed. The generated probability model effectively maps the dynamic change of the network state and provides a quantitative basis for risk assessment.

In the risk analysis submodule, the risk state and the change trend of the nodes in the network are analyzed through a probability model. And applying conditional probability query and structured query language of the Bayesian network to deeply analyze the risk condition and future development trend of each node in the electrical network. Based on the output of the dynamic Bayesian network, node risk states including fault and risk level assessment are refined, a risk state analysis result is generated, accurate information is provided for subsequent risk management and decision making, and identification of high risk areas and key risk factors is facilitated.

In the state updating sub-module, through a risk state analysis result, a real-time data flow analysis technology and an online learning mechanism of a dynamic Bayesian network are adopted to circularly adjust and update the risk assessment of the network. And collecting network operation data in real time, and adjusting a probability model through a dynamic Bayesian network by combining the existing risk analysis result to ensure that the model reflects the latest network state and risk level. And the generated real-time risk state result provides continuous risk monitoring and early warning service for the electrical network, and ensures the safety and stability of network operation.

Referring to fig. 2 and 7, the graph theory fault propagation analysis module includes a structure analysis sub-module, a risk estimation sub-module, and a path generation sub-module;

the risk prediction sub-module analyzes propagation paths of faults and risks in an electrical network by adopting a graph theory algorithm comprising network flow analysis and propagation dynamics model based on a network structure analysis result to generate a risk prediction result;

the path generation sub-module analyzes the fault occurrence area by referring to the topological structure of the electric network and real-time risk analysis through a path optimization algorithm based on the risk prediction result, and generates a fault prediction propagation path.

In the structural analysis submodule, a graph model is built on the basis of nodes and edges of an electrical network through an analysis method of graph theory. And quantitatively evaluating connectivity and influence of nodes in the network by applying network topology stability analysis and a node influence evaluation algorithm. Including calculating the node's degree, betweenness, and affinity centrality index to identify key nodes and potentially weak links in the network. The method helps to determine the core structure and weak links of the electrical network, generates a network structure analysis result, describes the topological structure and important nodes of the electrical network in detail, and provides a basis for risk prediction.

And in the risk prediction sub-module, deep analysis is carried out by adopting a graph theory algorithm through a network structure analysis result. Network flow analysis and propagation dynamics models are applied to predict propagation paths of faults and risks through network flows and node states. Traffic distribution, node load capacity, and fault prediction propagation paths in the network are analyzed to assess risk levels of the network. And finally, generating a risk prediction result, wherein the risk prediction result comprises a fault prediction propagation path and a high risk area, and providing key information for subsequent risk management and emergency preparation.

In the path generation sub-module, the electrical network is subjected to deep analysis by using a path optimization algorithm. And identifying the area where the fault occurs by referring to the topological structure of the electrical network and real-time risk analysis through nodes and edges in the network diagram. By implementing a shortest path algorithm or a minimum cost flow algorithm, the optimal path and critical propagation nodes for fault propagation are identified. The generated fault prediction propagation paths provide specific fault prediction and prevention strategies for the operation and maintenance of the electrical network, helping to reduce potential operation and maintenance costs and risk losses.

Referring to fig. 2 and 8, the time sequence analysis module includes a data sorting sub-module, a trend analysis sub-module, and a prediction model sub-module;

the trend analysis submodule analyzes periodic changes and trend directions in data based on the sorted data set by adopting seasonal decomposition and trend evaluation technologies, predicts load changes in a future time period and generates a trend analysis result;

the prediction model submodule is used for constructing a composite time sequence prediction model based on the trend analysis result and combining an autoregressive moving average model and a long-term and short-term memory network, predicting the change trend of the electrical load and generating a load prediction result.

In the data sort sub-module, the electrical equipment operation parameters and the environment monitoring data related to the fault prediction propagation path are processed through a data processing technology. And identifying and removing outliers in the data through outlier detection, processing missing values in the data by using an interpolation method, and performing standardized processing on the data by using a Z scoring method to ensure that the data has uniform measurement and range. The consolidated data set is generated to provide a clean, canonical, and reliable data base for trend analysis and prediction of future load changes.

In the trend analysis sub-module, seasonal, trend and random components in the time series are identified by seasonal decomposition and trend assessment techniques. The time series is decomposed into constituent parts using a time series decomposition method such as seasonal and trend decomposition using a Loess (STL) analysis method, and each part is independently analyzed to identify periodic variations and long-term trends in the data. The generated trend analysis results provide clear understanding of the future trend of the electrical load and assist in more accurate load management and planning.

In the prediction model sub-module, a composite time sequence prediction model is constructed by combining an autoregressive moving average model and a long-term and short-term memory network, and the model is used for accurately predicting the future change trend of the electrical load by combining the statistical analysis capability of an ARIMA model and the deep learning characteristic of an LSTM network. By adjusting model parameters, training the model and verifying the prediction accuracy thereof, a load prediction result is generated, so that accurate prediction of future loads is provided for the operation of the power system, and more effective energy distribution and load management strategies are facilitated to be formulated.

Referring to fig. 2 and 9, the flame monitoring module includes a dynamic flame analysis sub-module, a smoke behavior evaluation sub-module, and a scene discrimination sub-module;

The dynamic flame analysis submodule adopts a convolutional neural network algorithm to extract flame characteristics in the image based on a load prediction result, adopts a multi-layer filter to extract the edge, texture and color information of the flame, carries out hierarchical learning on the characteristics and generates flame characteristic identification data;

the smoke behavior evaluation sub-module is used for carrying out time sequence analysis on smoke dynamic behaviors by adopting a long-period memory network algorithm based on flame characteristic identification data, and generating a smoke behavior evaluation result by processing the change of smoke characteristics along with time through a neural network;

the scene discrimination sub-module adopts a random forest classification algorithm to discriminate fire scenes from non-fire scenes based on the smoke behavior evaluation result, and analyzes scene features by constructing a plurality of decision trees to generate fire analysis and recognition results.

In the dynamic flame analysis submodule, the environment around the electrical equipment is revealed through a convolutional neural network algorithm. The network automatically extracts key features of flames in the image, such as edges, textures and colors by using the multi-layer filter, and performs hierarchical feature learning. Training includes forward and reverse propagation, automatically adjusting network weights to minimize recognition errors. The generated flame characteristic identification data accurately describe the visual characteristics of the flame and provide accurate basis for smoke behavior evaluation and fire scene discrimination.

In the smoke behavior evaluation sub-module, the change of flame characteristics with time is displayed through a long-short-term memory network algorithm. The LSTM network is used to process time series data to identify dynamic patterns and periodic variations in smoke behavior. Network training involves learning behavioral characteristics of smoke using time series data, optimizing model parameters by iterative learning. And finally, generating a smoke behavior evaluation result, reflecting behavior characteristics of smoke developed along with time in detail, and providing key time sequence information for final fire scene discrimination.

In the scene discrimination sub-module, feature data obtained from dynamic flame analysis and smoke behavior evaluation are included through a random forest classification algorithm. And constructing a plurality of decision trees, and independently judging fire scenes and non-fire scenes according to the input feature vectors by each tree. And determining a final classification result through a voting mechanism. And the generated fire analysis and recognition result can accurately distinguish fire scenes from non-fire scenes and provide effective decision support for fire early warning and emergency response.

Referring to fig. 10, the fire monitoring method for an electrical device, which is performed based on the fire monitoring system for an electrical device, includes the steps of:

s3: based on fault diagnosis results, analyzing the positions and functions of nodes of the electrical equipment in a network by adopting a network analysis method, simulating the functions of the nodes in the electrical network, determining the diffusion paths and risk areas of fire and faults by adopting a graph theory algorithm, and generating a fire spreading model;

s4: based on a fire propagation model, a dynamic Bayesian network is adopted, a model is built by combining a particle filtering algorithm, real-time probability of a network state is estimated and updated, risk analysis is carried out by combining historical data, risk areas and change trends in the network are identified and predicted, and a real-time risk state is generated;

s6: based on a network structure analysis result, adopting a graph theory algorithm, combining network flow analysis and a propagation dynamics model, and carrying out predictive analysis on propagation paths of faults and risks in an electrical network to generate a fault predictive propagation path;

s7: based on a fault prediction propagation path, analyzing periodic changes in the path by adopting a seasonal decomposition technology, determining the trend direction of the path by using a trend evaluation technology, constructing a composite time sequence prediction model by combining an autoregressive moving average model, predicting the electrical load change trend in a future time period, and generating a load prediction result;

s8: based on the load prediction result, a convolutional neural network algorithm is adopted to process images around the electrical equipment, flame characteristic identification data are analyzed, dynamic behavior of smoke is analyzed, a random forest classification algorithm is adopted to analyze real-time scenes, fire conditions are judged and identified, and fire analysis and identification results are generated.

In step S1, images of smoke and flames in the surrounding environment are acquired by dynamic range imaging and image enhancement techniques. Preprocessing, including denoising, contrast enhancement, and brightness adjustment, is performed via image processing algorithms to ensure that flame and smoke characteristics are effectively identified. The characteristic extraction is carried out by adopting a convolutional neural network, the network structure design comprises a plurality of convolutional layers and pooled layers, and key characteristics such as shape, color and texture in the image are extracted layer by layer. The deep learning model can identify and distinguish the differentiation stage and type of flame and smoke through the training of the batch image data. Through iterative learning, the model is circularly optimized, and the accuracy of flame and smoke identification is improved. The risk assessment results are generated, and the fire risk level is determined by analyzing image characteristics such as flame size, smoke density and diffusion speed. The evaluation results are important for quick response and taking corresponding fire precautions.

In step S2, key operating parameters and monitoring data are collected from the electrical equipment and the surrounding environment using real-time data stream acquisition techniques and data processing techniques. The collected data is analyzed using anomaly detection algorithms, such as isolated forests or gaussian mixture models, to identify data points that differ significantly from the normal operating mode, i.e., anomaly modes. Anomalies represent potential equipment failures or abnormal operating conditions, which pose a risk to electrical safety. The data is cleaned, normalized and feature selected to ensure accuracy of the analysis. The generated fault diagnosis results include specific fault types, affected equipment and recommended maintenance measures, which are critical for maintaining safe operation of the electrical equipment and preventing occurrence of fire.

In step S3, based on the fault diagnosis result, a network analysis method and a graph theory algorithm are adopted, including nodes and edges of the network. The network analysis method is applied to an electrical network to identify and evaluate the location and function of nodes in the network. Algorithms such as centrality analysis are used to determine key nodes in the network, which are critical to the overall operation of the network. Centrality analysis includes differentiated forms of centrality, median centrality, tight centrality, and the like, each revealing the importance of a node from a differentiation perspective. Graph theory algorithms are used to model the role of nodes in an electrical network and determine the propagation paths of faults and fires. A network model is built using, for example, a shortest path algorithm, the max-flow min-cut theorem, to predict the path of fault propagation. The topology of the network, the strength of the connections between nodes, and the mechanism of fault propagation are considered. The generated fire propagation model details graphically or in report how a fault propagates in the network in a particular fault situation, which nodes and paths will be affected, and potentially risk areas. The network operators are helped to optimize the network design, the protection of key equipment is enhanced, the influence of potential faults is improved, the stable and safe operation of the network is ensured, and the reliability and the efficiency of the electrical network are obviously improved.

In step S4, a real-time probability model of the network state is established and updated by using a dynamic Bayesian network and a particle filtering algorithm. And continuously evaluating the risk state in the network by calculating the conditional probability by combining the historical data and the real-time monitoring information. Including quantifying abnormal states of network states, estimating and updating states using particle filtering to reflect the latest network states and risk levels. The generated real-time risk status provides a view of real-time updates for risk management and decision making in the network, making risk preventive measures more targeted and time-efficient, thereby reducing potential risks and losses.

In step S5, the structural data of the electrical network, including node and connection line information, is processed by a graph theory method, reflecting the interconnection network structure between the devices. And analyzing and evaluating the stability of the nodes by the network topology stability, and identifying the network weakness by using the connection strength among the nodes and the overall network connectivity index. And (5) evaluating the influence of the nodes, and identifying key nodes according to indexes such as degree centrality, medium centrality and the like. Network analysis software or algorithms are used to perform centrality analysis, aided by network simulation and sensitivity analysis to reveal network dynamics and potential risks. And the generated network structure analysis result details the key part and the fragile link of the electrical network and provides an improvement scheme. The network structure, key nodes and potential risks are comprehensively understood, the network structure is very important for optimizing network operation, improving stability and safety and making maintenance and upgrading plans, faults can be effectively prevented, and economic and safety risks are reduced.

In step S6, the electrical network graphic data composed of the electrical equipment and the connection lines is processed by using graph theory algorithm, network flow analysis and propagation dynamics model. And simulating the propagation mode of the fault in the power grid by using network flow analysis, calculating the maximum flow and potential bottleneck under the differential path, and predicting the fault propagation path. The propagation dynamics model simulates the dynamic propagation process of the fault in the network, evaluates the influence of the differentiated fault types and propagation speed and range, including network topology, node importance and fault propagation characteristics. Complex data and algorithms are processed using high-level programming languages and network analysis tools such as Matlab or Python's network x. And predicting the propagation path, speed and influence area of the fault in the network by simulating the differential fault condition, and finally forming a fault prediction propagation path. The fault propagation paths, affected nodes and risk areas are detailed, providing critical information for risk mitigation strategies and emergency plans. And preventive measures are convenient to take for specific weak points or key nodes, so that the influence of faults is effectively reduced, and the stability and safety of the network are ensured. Not only is the understanding of the power grid structure and potential fault paths enhanced, but also basis is provided for scientific formulation of risk management and emergency response, the fault processing efficiency is improved, and the loss caused by fault diffusion is remarkably reduced.

In step S7, the time series data is decomposed into trend, seasonal and residual components using seasonal decomposition techniques, such as seasonal and trend decomposition using the Loess (STL) method, by means of a seasonal decomposition technique and an autoregressive moving average model. Periodic changes in the data are identified, providing a clear baseline for trend assessment. The long-term trend direction of the data is determined using trend estimation techniques, such as moving average or horrick-Prescott filters. A composite time series prediction model is constructed in conjunction with an autoregressive moving average (ARMA) model or a variant thereof, such as a seasonal autoregressive integrated moving average model (SARIMA). Including determining parameters of the model such as autoregressive terms, differential order and moving average terms, and seasonal components. Model fitting is performed through historical data, statistical tests such as AIC or BIC are used to select an optimal model, and future electrical load change trends are predicted by using the model. The generated load prediction result provides scientific basis for operation and maintenance decisions of the electrical network, and the prediction result can guide how to adjust power supply so as to meet expected load change, reduce fault risk, optimize resource allocation and improve energy efficiency and economic benefit. The change trend of the load demand can be better understood and predicted, so that more intelligent decisions can be made in the aspects of maintenance planning, load management, emergency preparation and the like. By reducing unexpected power outages and optimizing network operation, quality of service and customer satisfaction are improved while operating costs and environmental impact are reduced.

In step S8, abnormal conditions in the electrical environment, such as flames and smoke, are captured by a Convolutional Neural Network (CNN) algorithm and a random forest classification algorithm. The convolutional neural network is adopted to process the image, and the network automatically extracts key characteristics of flame and smoke, such as shape, color, texture and the like through a plurality of convolutional layers and a pooling layer. The image data is preprocessed, including size adjustment, normalization and image contrast enhancement, so that the model training efficiency and accuracy are improved. A CNN model structure is defined, and an appropriate activation function, such as ReLU, and an appropriate loss function, such as cross entropy, are selected. The model is trained by training the data set, and the network weights are optimized by using a back propagation algorithm, so that the performance of the model in flame and smoke recognition is enhanced. After multiple iterations, the CNN model is able to identify and distinguish flame and smoke features in the surrounding environment of the electrical device. And carrying out iterative analysis by adopting a random forest classification algorithm based on flame characteristic identification data output by the CNN model. The method comprises the steps of constructing a plurality of decision trees, independently making judgment on the basis of flame characteristic data by each tree, and determining a final classification result through a voting mechanism. The generated fire analysis and recognition result details the real-time state of the surrounding environment of the electrical equipment, including the presence of fire and the location, size and diffusion rate of the flames and smoke. The information is important to timely respond to the fire, make an emergency evacuation plan and start fire-fighting measures, and is helpful to improve the loss and influence caused by the fire.

The present invention is not limited to the above embodiments, and any equivalent embodiments which are changed or modified to equivalent changes by those skilled in the art can be applied to other fields by using the technical contents disclosed above, but any simple modification, equivalent changes and modification made to the above embodiments according to the technical matters of the present invention are still within the scope of the technical solutions of the present invention.

Claims

1. A fire monitoring system for electrical equipment, its characterized in that: the system comprises a biological characteristic recognition module, a space-time data analysis module, a network theory module, a dynamic Bayesian network module, a graph theory fault propagation analysis module, a time sequence analysis module and a flame monitoring module;

2. The fire monitoring system for electrical equipment of claim 1, wherein: the risk assessment results comprise probability scores of fire occurrence, fire type identification and region positioning of fire occurrence, the fault diagnosis results comprise types of faults of electrical equipment, time point predictions of faults and space positions of faults, the fire propagation model comprises paths of fire propagation, network node scores of fire influence and definition of risk regions, the real-time risk state comprises fire risk assessment values updated based on real-time data, multiple types of risk factors and predicted risk trends, the fault prediction propagation paths comprise electrical network branches of fault influence, speed of fault diffusion and identification of control nodes, the load prediction results comprise trend predictions of loads in future time periods of the electrical equipment, time intervals of load fluctuation and predicted load peaks, and the fire analysis and identification results comprise flame positions, flame sizes and smoke densities.

3. The fire monitoring system for electrical equipment of claim 1, wherein: the biological characteristic recognition module comprises an image acquisition sub-module, a characteristic analysis sub-module and a risk assessment sub-module;

4. The fire monitoring system for electrical equipment of claim 1, wherein: the space-time data analysis module comprises a data acquisition sub-module, a model analysis sub-module and a diagnosis result sub-module;

5. The fire monitoring system for electrical equipment of claim 1, wherein: the network theory module comprises a node analysis sub-module, a dynamics simulation sub-module and a path identification sub-module;

6. The fire monitoring system for electrical equipment of claim 1, wherein: the dynamic Bayesian network module comprises a probability building sub-module, a risk analysis sub-module and a state updating sub-module;

7. The fire monitoring system for electrical equipment of claim 1, wherein: the graph theory fault propagation analysis module comprises a structure analysis sub-module, a risk prediction sub-module and a path generation sub-module;

8. The fire monitoring system for electrical equipment of claim 1, wherein: the time sequence analysis module comprises a data arrangement sub-module, a trend analysis sub-module and a prediction model sub-module;

9. The fire monitoring system for electrical equipment of claim 1, wherein: the flame monitoring module comprises a dynamic flame analysis sub-module, a smoke behavior evaluation sub-module and a scene discrimination sub-module;

10. Fire monitoring method for electrical equipment, characterized in that the fire monitoring system for electrical equipment according to any of claims 1-9 is performed comprising the steps of:

based on the surrounding environment of the electrical equipment, adopting dynamic range imaging and image enhancement technology to collect images of smoke and flame in the environment, extracting key features in the images, iteratively identifying morphological features of the smoke and the flame, evaluating fire risks, determining risk grades by analyzing the image features, and generating a risk evaluation result;

based on the risk assessment result, collecting and arranging operation parameters and environment monitoring data of the electrical equipment by adopting a real-time data flow acquisition technology and a data processing technology, analyzing the operation condition and environment change trend of the electrical equipment, performing fault diagnosis on the electrical equipment by using an anomaly detection algorithm, identifying an anomaly mode in the data, and generating a fault diagnosis result;

Based on the fault diagnosis result, analyzing the node position and function of the electrical equipment in the network by adopting a network analysis method, simulating the function of the nodes in the electrical network, determining the diffusion path and risk area of fire and fault by adopting a graph theory algorithm, and generating a fire spreading model;

based on the fire spreading model, a dynamic Bayesian network is adopted, a probability model is built by combining a particle filtering algorithm, the real-time probability of the network state is estimated and updated, risk analysis is carried out by combining historical data, risk areas and change trends in the network are identified and predicted, and a real-time risk state is generated;

based on the real-time risk state, adopting a graph theory analysis method, including network topology stability analysis and node influence evaluation, identifying key nodes and links in a network, and generating a network structure analysis result;

based on the network structure analysis result, adopting a graph theory algorithm, combining network flow analysis and a propagation dynamics model, and carrying out predictive analysis on propagation paths of faults and risks in an electrical network to generate a fault predictive propagation path;

based on the fault prediction propagation path, analyzing periodic changes in the path by adopting a seasonal decomposition technology, determining the trend direction of the path by using a trend evaluation technology, constructing a composite time sequence prediction model by combining an autoregressive moving average model, predicting the electrical load change trend in a future time period, and generating a load prediction result;

Based on the load prediction result, a convolutional neural network algorithm is adopted to process images around the electrical equipment, flame characteristic identification data are analyzed, dynamic behavior of smoke is analyzed, a random forest classification algorithm is adopted to analyze a real-time scene, fire conditions are judged and identified, and fire analysis and identification results are generated.