A fire identification method for power equipment based on multi-parameter recursive graph

Abstract

The invention provides a method for identifying fires in electric equipment based on a multi-parameter recursive graph, and relates to the technical field of electrical fire detection. The method comprises: using a smoothing filtering technology to pre-process monitoring video images, and creating a fire identification algorithm model based on artificial intelligence to detect and identify fire images in suspicious areas; the fire identification algorithm model identifies the hidden danger electrical monitoring data of the detection position, calculates the target feature vector corresponding to the trace features of the target video image, and builds a deep neural network sub-model and a power heterogeneous multi-parameter fusion model; then using a recursive algorithm module to arrange the power time series parameters into a recursive matrix, and constructing a graph convolutional neural network based on a multi-parameter recursive graph to extract features; using a fire warning judgment module to segment the suspicious unknown areas of the image sequence where the fire occurs, and timely predict and process them, and after identification and confirmation, automatically start a multi-parameter warning judgment program, and accelerate the issuance of warning signals by a background warning system.

Description

Technical Field

[0001] The invention relates to the technical field of electrical fire detection, and in particular to a method for identifying fires of electric equipment based on a multi-parameter recursive graph. Background Art

[0002] Although a very small number of insulating materials of electrical equipment or wires are inorganic materials, most of them are flammable organic insulating materials. When the line current is too large, according to the principle of conservation of energy, a large amount of electrical energy is converted into heat energy, which causes local overheating of the conductor and insulating material. When a certain temperature is reached, a fire will occur. Under normal power-on conditions, all normally operating electrical equipment will generate heat, which means that it is a potential source of fire or heat. When electrical equipment operates abnormally or is powered on for a long time without supervision and management, the large amount of heat generated may cause the temperature in the environment to reach the ignition point of nearby flammable materials, leading to the occurrence of fire. Static electricity may also cause static electricity fires. Static electricity fires are often caused by contact, separation, and mutual friction between different objects. When static electricity increases to a certain level, under certain conditions, it can break down the air medium. In this case, spark discharge often occurs. Spark discharge is similar to an electric arc and will generate a large amount of heat energy, which is likely to cause nearby combustibles or explosive mixtures to burn or explode. Overheating is a major form of electrical fire. If the insulation layer of electrical equipment is aged or damaged, it will cause excessive loss of the insulating medium. Under the action of the operating voltage, it will overheat and become one of the hidden dangers of electrical fires. Short circuit is one of the most serious fault conditions in electrical equipment. Electrical short circuit is caused by the long-term use of electrical equipment, the aging, damage and breakdown of the insulation layer, or the loss of insulation due to high temperature, humidity, corrosion and other effects, resulting in the collision of two wires. According to measurements, an arc current of 0.5A may produce an arc temperature of thousands of degrees Celsius, which can not only ignite its own insulating materials, but also ignite the combustible materials, steam and dust nearby. Statistics show that electrical fires caused by short circuits account for 30% to 45% of all electrical fires. Electrical fires are generally caused by short circuit faults, overload faults, insulation faults, contact resistance faults, sine wave distortion faults and other reasons. The phenomena caused by these faults are mainly high temperature, arc, ultrasound, ultraviolet, smoke, and leakage current. When leakage current is generated, it is usually not directly measured because the leakage position cannot be determined. From the analysis, it can be seen that resistive fault arcs can cause harmonic distortion. The waveform distortion in the power system is mainly generated by power equipment, which inevitably causes a large degree of distortion in the related voltage and current waveforms, which is commonly known as harmonic pollution. The harmonic pollution caused by these nonlinear devices also causes certain interference to the identification of arc faults. According to the more than 970,000 fire accidents investigated by fire departments in the past ten years, there were 240,000 electrical fires, accounting for as high as 24.7%, and the economic losses caused by electrical fires accounted for 36.9% of the total fire losses. Traditional fire smoke and temperature detection equipment has low sensitivity and delayed response in such scenarios, and it is difficult to meet the battery thermal runaway warning needs. At this stage, the intelligent fire detection system is not mature enough, and it is difficult to accurately collect static and dynamic visual representation data of fires.

[0003] In the application process based on image processing, the video surveillance system may have great difficulties in the process of detecting external factors such as position interference due to unstable external environmental lighting conditions and signal interference from the camera system, and collect monitoring videos with low image quality, making the image details in the video blurred and difficult to be accurately analyzed and identified by subsequent intelligent machines, reducing the efficiency of fire detection and identification. In the actual application of intelligent fire detection algorithms based on image processing at this stage, many key technical difficulties have not been effectively solved, and it is difficult to intelligently select appropriate fire judgment standards according to different environmental occasions, making it difficult to effectively improve the accuracy of fire detection and identification. Firefighting construction has not yet been fully developed in the vertical high-altitude field, which leads to great difficulties and challenges in high-rise building fire fighting.

[0004] In recent years, the development trend of the construction industry has been particularly rapid. The height of buildings has been increasing, and the space area has also increased accordingly. In addition, the spatial structure of urban high-rise buildings is complex. A large number of corridors, passages, stairs, elevators and other facilities have various fire safety hazards, which has led to an increase in the probability of fires in urban building spaces, making it difficult to carry out fire fighting work smoothly. Among them, building electrical majors account for a considerable proportion of buildings. Today's buildings, in terms of architectural design, pursue different shapes and beautiful appearances. Indoor and outdoor decoration and electrical line laying are even more complicated, causing fires caused by electrical lines to occur frequently and are increasing year by year. Traditional fire detection and identification technology can no longer meet the needs of accurate detection and identification of fires in complex building spaces. Considering the different seasons and temperature fluctuations during the day, the change in external temperature is generally always slow, and the temperature change of the power supply line caused by electrical faults is often relatively fast. The occurrence of electrical fires is usually caused by abnormally high temperatures caused by electrical faults. Therefore, the temperature change rate and the maximum allowable operating temperature can usually be used to reflect whether an electrical fire may occur in the power supply line and the impact of the external ambient temperature on the detection. The occurrence of some electrical faults is often accompanied by the appearance of electric arcs, and the temperature of electricity can reach 2000℃-3000℃, which is very likely to cause electrical fires. Especially when the transient voltage is superimposed on the operating voltage, the high-amplitude transient impulse voltage will have an excitation effect on the defects, excite the generation of partial discharge, and generate continuous partial discharge under the subsequent continuous operating voltage, which will lead to the occurrence of accidents. Experiments have shown that there are many electrical fault characteristic parameters that can be used for thermal runaway warning, and each parameter has differences in the timeliness, reliability and implementation difficulty of the warning. Since the process of fire accidents is accompanied by physical and chemical phenomena such as smoke, flames, and noise, the fire detection and identification parameters that are currently widely used include smoke, flames, noise, images and other characteristic parameter linkages. Smoke contains gaseous substances and liquid-solid substances produced by combustion. Its temperature, concentration and gas phase characteristic spectrum are easy-to-detect parameter characteristics, which can identify the occurrence of fire according to the changes in parameters; but traditional fire smoke and temperature detection equipment have low sensitivity and delayed response in such scenarios, and it is difficult to meet the battery thermal runaway warning needs.

[0005] Because electrical fires are sudden and hidden, especially some electrical projects are not installed in a formal way and the wiring is chaotic, which can easily cause electrical faults and fires. Specifically, electrical fire hazards can be divided into two types: overheating hazards and discharge hazards. Overheating hazards refer to the abnormal increase in temperature of the heating part of the electrical equipment or line, which ignites the insulation layer of the electrical equipment and line, and then ignites the surrounding combustibles, causing fires; while discharge hazards refer to the insulation layer of the electrical equipment and line itself being damaged, and the electric energy is released, which causes free corona on the surface, which can easily cause insulation breakdown, specifically manifested as electric sparks and arcs. When a short circuit or grounding fault occurs, electrical discharge will generate electricity, and the arc will emit 25~45kHz ultrasound, which can be used to detect the discharge arc. Through the analysis of common electrical faults, it is concluded that the generation of arcs and high temperatures is often caused by abnormal leakage current of the distribution line. When the distribution line and the electrical equipment in the line have problems such as aging, damage, water ingress, and insulation degradation caused by non-standard construction, the residual current in the entire line will increase. Purely resistive leakage current in the line can be considered as leakage current caused by reduced insulation. Generally, residual current measurement can be used to reflect the existence of abnormal leakage current. Selecting appropriate residual current test points can determine whether there is abnormal leakage current in the system. This is much easier than directly testing the leakage current. After all, the location of leakage current is usually unpredictable. However, existing electrical fire monitoring equipment only independently detects residual current, fault arc, temperature, pyrolytic particles and other information. It is difficult to conduct comprehensive and effective detection, analysis and judgment of specific circuits of the distribution circuit. Therefore, the accuracy and efficiency of detecting electrical fires are not high. Residual current electrical fire monitoring detectors are mainly used in the normal working state of distribution lines to promptly detect faults and fire hazards in the lines.

[0006] At present, ice perception of transmission lines can be divided into two categories: based on sensor data and based on image data, that is, ice perception is performed using sensor data or image data alone as input. The state of the power perception target at a certain moment is the result of the joint action of multiple parameters over a period of time, so it has time and space characteristics. Although heterogeneous multi-parameter fusion takes multiple parameters with different structural forms and physical meanings as input, it is difficult to make a unified description due to the different representation spaces of each parameter, and the fusion is difficult. In addition, the cross-type and multi-dimensional data analysis technology is weak, and the correlation analysis and mining capabilities between state quantities are insufficient. The contradiction between the urgent need for heterogeneous multi-parameter fusion and the limited technical means is still prominent.

[0007] The existing electrical fire monitoring system belongs to the early warning system for leakage monitoring. The structural forms of its monitoring detectors are mainly three types: multi-function leakage switch type, separate configuration type, and separate configuration integrated type:

[0008] The structure of the multifunctional leakage switch is boxed from the outside, and from the inside, it integrates the monitoring detector containing power conversion circuit, signal processing circuit, alarm circuit, display circuit, communication, linkage interface, etc. with current transformer, residual current transformer, and main circuit disconnect switch to form a multifunctional leakage switch product. The advantages of this type of product are: multiple protection functions, less wiring, and high integrity. The disadvantages of this type of built-in current transformer and residual current transformer (including temperature sensor) are: complex structure, increased cost, and high failure rate. In particular, the distance between the signal monitoring, detection, analysis, processing, alarm, communication, linkage interface and other circuits and the ABC three-phase main power (or single-phase) circuit is too close, which is easily interfered by strong electromagnetic fields, reducing the stability and reliability of product performance. Besides, it is not very convenient to install and use;

[0009] The separate configuration integrated type is a special type of electrical fire monitoring equipment and electrical fire monitoring detectors. It is easy to manage. The trouble is that the relatively complex fire alarm system has a high failure rate, which directly affects the electrical fire monitoring system, reduces the stability and reliability of the system, and even causes paralysis.

[0010] The separation configuration type is the electrical fire monitoring equipment and the electrical fire monitoring detector. This system has a clear division of labor, simple structure, low cost, low failure rate, and stable and reliable performance. The disadvantage is that signal lines and 2-core tripping control lines need to be laid between the monitoring equipment and the monitoring detector, and between the monitoring detector and the terminal detection head.

[0011] In the field of fire detection, most traditional fire image detection algorithms use flames as detection targets. However, smoke is generated at the beginning of a fire, and flames are generated only in the middle of a fire, which makes it impossible to prevent and control the fire in the first place. At the same time, the common algorithm for identifying smoke with a single feature has a lower recognition rate and a sharp increase in the case of complex scenes or more interference. The existing algorithm has a low accuracy rate for the foreground of flames, resulting in poor overall robustness of the system. Whether it is flame detection or smoke detection, there is a problem that the detection rate and the false detection rate cannot be balanced, that is, when the detection rate is high, the false detection rate will also increase, and vice versa. The algorithms that meet the requirements for both the false detection rate and the detection rate often have high computational complexity and cannot meet the needs of real-time detection. In the field of flame detection, the detection rate of long-distance flames is low, the moving objects of flame color cause a high false detection rate, and the computational complexity is high. Most existing algorithms make hard decisions in the foreground detection stage and the regional analysis stage, and cascade these two stages, resulting in erroneous results in any stage affecting the final detection. Summary of the invention

[0012] The main purpose of the present invention is to provide a method for identifying fires in power equipment based on a multi-parameter recursive graph, so as to solve the technical problems of low fire identification accuracy and poor robustness in the prior art.

[0013] In order to achieve the above-mentioned purpose, according to one aspect of the present invention, a method for identifying fires in power equipment based on a multi-parameter recursive graph is provided, comprising: using an intelligent fire detection algorithm module to obtain electrical monitoring data of a target location and a target video image to be detected, wherein the target location includes the location of a monitoring area of ​​an intelligent monitoring system for fire early warning of power equipment, integrating electrical monitoring data and target video images to obtain power image parameters, and creating a fire identification algorithm model based on artificial intelligence based on the power image parameters; the fire identification algorithm model monitors video image segmentation and regional detection, identifies the hidden danger electrical monitoring data of the target location, builds a deep neural network sub-model, extracts features from the power image parameters based on the deep neural network sub-model, obtains target features, and creates a power heterogeneous multi-parameter fusion model based on the target features; based on the power heterogeneous multi-parameter fusion model, data-level fusion is performed on power image parameters and power timing parameters to obtain power perception data, feature extraction is performed on the power heterogeneous multi-parameter fusion model, and the power timing parameters are arranged into a k×n recursive matrix based on the recursive algorithm module, wherein, k is the number of categories of the time series, n is the length of the time series, and a multi-parameter recursive graph is obtained based on the recursive matrix. A graph convolutional neural network is constructed based on the multi-parameter recursive graph to extract features. The fire warning judgment module is used to segment the suspicious unknown areas of the image sequence where the fire occurred for pre-judgment processing, and the judgment results of each feature are fused. After identification and confirmation, it is determined whether the fire has reached the alarm level. When a warning safety signal is detected, the multi-parameter warning judgment program is automatically started, and the background warning system accelerates the issuance of warning signals.

[0014] Optionally, the power equipment fire identification method adopts a power equipment fire identification device for fire identification, and the power equipment fire identification device includes: an intelligent fire detection algorithm module associated with the residual current electrical fire monitoring system, a recursive algorithm module, a fire warning judgment module and a residual current fire alarm module formed by a residual current electrical fire monitoring detector, a temperature measurement electrical fire monitoring detector, and multiple bus isolators connected via a dual bus, wherein the residual current electrical fire monitoring system adopts a hierarchical distributed structure, when the ground fault current detected by the residual current fire alarm module in the protected line exceeds a preset value, after analysis and confirmation, an alarm signal and a control signal are issued, and at the same time, the ground fault signal is transmitted to the main control machine through the bus within a few seconds, the main control machine sends an alarm signal and a control signal, the display screen displays the alarm address, and the alarm and control information are recorded and saved.

[0015] Optionally, an intelligent fire detection algorithm module is obtained based on flame data recognition training. The intelligent fire detection algorithm module cooperates with a camera to obtain a target video image of a target position in real time, and uses preprocessing technology to eliminate interference factors in the target video image to highlight the effective information body of the target video image.

[0016] Optionally, based on the intelligent fire detection algorithm module, the parameter data of the area unit covered by the target video image is obtained, the target video image is segmented into similar areas based on the video image segmentation algorithm and identification and detection are performed, and the segmentation threshold of the flame image is calculated based on the maximum inter-class variance algorithm and segmentation processing is performed; for the joint image segmentation of the space domain and time domain, the Gaussian mixture model is used to detect the flame target, the Gaussian mixture model applies a weighted combination of multiple unimodal distribution functions to approximate any probability distribution, estimates the parameters and weights of each mixed component, and uses the morphology-based flame binary image post-processing technology to filter image noise.

[0017] Optionally, based on the fire picture set collected by the intelligent fire detection module, the fire recognition algorithm model is trained and identified using an artificial intelligence perception information processing algorithm and a deep learning fire smoke detection algorithm, and a large number of image learning classifiers are trained; based on the convolution layer, different features of the target video image are identified and extracted, and based on the fuzzy neural network algorithm, the feature representation and relationship of the target video image are learned, and fire characteristics are learned and trained to automatically generate membership function parameters suitable for the fire detection model; after the training, the weights of the relative effects of each fuzzy inference rule are extracted, and the mapping rules between the fire parameters and the sensor input patterns are obtained through the changes in the weights, and the target video images are classified based on the mapping rules, and the current state is associated with the previous state using a recurrent unit, and each element in the sequence is processed, the current element is predicted, the next element of a sequence is predicted, and the next value of a time series is predicted, and a deep neural network sub-model is generated according to the training data.

[0018] Optionally, the fire identification algorithm model is based on a flame detection algorithm of wavelet transform and fuzzy support vector machine (FSVM), performs a two-dimensional wavelet transform on the target video image, extracts the square sum of wavelet coefficients of several frames of pixel points as eigenvalues, and applies the fuzzy support vector machine as a classifier to determine whether the extracted pixel points are flame area pixels. If they are, the extracted fusion features are used to train the support vector machine, and the extracted motion areas are classified into smoke and non-smoke through the classifier in turn. After detecting the flame or smoke, the detected position information and category information are sent to the deep neural network sub-model, and the fuzzy neural network is established with the smoke concentration and the smoke concentration change rate signal as input, and finally connected to the classification or regression layer for target perception.

[0019] Optionally, the fire recognition algorithm model first extracts flame features in the transform domain, calculates the time series of the farthest distance of the edge of the candidate flame area, and performs a two-dimensional Fourier transform on the sequence to obtain the frequency change characteristics, and inputs the R value of each candidate pixel in the RGB color space or the Y value in the YUV color space into the corresponding level filter for time wavelet analysis, and selects to combine morphological operations in the YUV color space to obtain the shadow pixel point set and the foreground target pixel point set, and the number of zero crossing points of the output wavelet sub-signal is used to distinguish whether it is a flame pixel point. At the same time, the mean of the spatial wavelet transform is compared with a fixed threshold, and the comparison result is also used as one of the bases for judging whether a flame exists.

[0020] Optionally, in the scenario of intelligent fire identification and monitoring, the fire warning judgment module performs intelligent judgment based on a multi-feature fusion smoke detection algorithm, using an artificial neural network as a fire identification algorithm model that can simulate the brain's thinking mode. The characterization parameters of the flame image segmentation area are used as input values, and the artificial neural network algorithm model is used to quickly eliminate environmental interference factors, automatically adjust operating parameters according to on-site changes, detect, identify and judge fires, and compare and analyze them with fire occurrence standards to obtain comparison results, determine the situation of the actual fire point based on the comparison results, and issue a warning in time.

[0021] Optionally, in the monitored scene, the fire warning judgment module first uses a brightness algorithm to calculate the brightness of the original image without flames, and sets the brightness of the original image without flames as a brightness threshold, and compares the set brightness threshold with the brightness value calculated from images of other scenes. If the brightness value calculated from the images of other scenes is greater than the brightness threshold, a fire may have occurred; if the brightness value calculated from the images of other scenes is less than the brightness threshold, no fire has occurred. When processing smoke images and flame images, the collected images are grayed out, and after pixel value analysis and processing, if the average brightness value obtained is higher than the detection threshold, it can be determined that there is a fire in the image, and an alarm window pops up for prompting; if it is lower than the detection threshold, there is no fire in the image.

[0022] According to another aspect of the present invention, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the above method is executed when the program is run.

[0023] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored executable program, wherein when the executable program is running, the device where the storage medium is located is controlled to execute the above method.

[0024] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, and the computer program implements the above method when executed by a processor.

[0025] The technical solution of the present invention is applied. The present invention aims at the current problems of frequent occurrence, rapid spread, large hazard coefficient and difficulty in prediction of electrical fires. Through the fire identification device of power equipment, the intelligent fire detection algorithm module is used for remote real-time monitoring, and the electrical monitoring data of the source position of the monitoring area of ​​the intelligent monitoring system for early warning of fire of power equipment and the target video image to be detected are obtained. The pixels with flame characteristics are screened out, and the parameters closely related to the fire condition are integrated. With the help of the thermal runaway process, the color, texture, size and shape of the flame are analyzed. With strong anti-interference ability and high resolution ability, the intelligent fire detection algorithm technology based on image processing can be remotely monitored in real time, and the target video image generated by the fire of the electrical circuit of the faulty power equipment can be identified. The operation is simple, the data storage volume is small, it is almost unaffected by the external environment, and the detection effect is stable. The test results show that the intelligent fire detection algorithm can accurately determine the location of the fire by using the principle of image vision, and the detection results are better than other algorithms in terms of detection rate and false detection rate. And the intelligent fire detection algorithm technology based on image processing can improve the speed and accuracy of fire detection. It can make intelligent analysis and adjustments based on the lighting conditions, airflow status, detection position and other characteristics of the building's interior space, ensure the accuracy of the detection and identification functions in different scene environments is not interfered with, filter out on-site factors that have little to do with the fire, and integrate parameters closely related to the fire situation for monitoring, so that power equipment of different types of structures and different purposes can use intelligent fire detection algorithm technology to timely predict and deal with fires.

[0026] The present invention uses smoothing filtering technology to pre-process the monitoring video image, remove image noise pollution, and eliminate interference factors in the image. While ensuring that the detection rate is not reduced compared with the existing algorithm, the false detection rate caused by interference is greatly reduced, and the calculation complexity is reduced. Most of the background pixels can be filtered out at this stage, which can reduce the amount of calculation for subsequent processing and avoid false detections that may be caused by background areas. In addition, by performing intelligent detection and sensing on the video image, the occurrence of fire can be intelligently sensed without physical contact, and the fire information can be intuitively presented. The fire detection system can also be added to the existing monitoring system, which can reduce the burden on the fire prevention system, improve the functionality of the system, and reduce costs. When applied to a larger space, the presence of a fire can be accurately and quickly detected.

[0027] The present invention aims at the problem that the single feature used in the fire image feature extraction method causes a high false alarm rate of fire. The fire recognition algorithm model is used to monitor the segmentation and regional detection of video images, identify the hidden danger electrical monitoring data of the detection position, and calculate the target feature vector corresponding to the trace feature of the target video image. The experimental results show that the fire recognition algorithm model has better stability, faster computing speed and higher recognition rate. Under the condition of minimum error probability, the result after recognition is made to conform to the actual situation to the greatest extent, and the accuracy of recognition is improved, so as to achieve the purpose of automatic classification and recognition of physical objects by machines. The target feature vector corresponding to the calculated trace feature of the target video image can send out an early warning signal after the gas sensor detects the trace feature gas production or the optical sensor captures the weak vaporized electrolyte image. The fire recognition algorithm model is combined with the fire detection algorithm to build a deep neural network sub-model. The sub-model extracts features of various parameters according to the data form and characteristics, and then processes the multivariate fire information data through the fire recognition algorithm model, which can improve the speed and accuracy of fire recognition and enhance the reliability of fire recognition. With the help of the deep neural network sub-model, the problem that the current intelligent fire detection algorithm is not adaptable enough to the external environment of the fire, and there are great difficulties in filtering the interference of various external factors such as lighting conditions, airflow conditions, and detection positions inside the building, and it is difficult to intelligently select appropriate fire judgment standards according to different environmental occasions, making it difficult to effectively improve the accuracy of fire detection and recognition. After neural network training, the actual recognition accuracy rate exceeds 98.1%, and the recognition accuracy rate of optical image judgment criteria exceeds 90%. Combined with characteristic gas recognition, the system reliability is better than the current single-criteria thermal runaway warning method. Extracting key smoke dynamic features and various features of texture feature target images from the recursive graph can accurately identify the basic features of flame and smoke images, reduce the influence of other interference objects on fire judgment, accurately identify fires, and improve the detection accuracy of the system. Detecting pixels with changed brightness or grayscale values ​​of smoke dynamic features through the foreground detection stage is conducive to improving the effectiveness of fire recognition. Constructing a heterogeneous multi-parameter fusion model for electric power can make heterogeneous parameters complement and enhance each other, fully explore the nonlinear relationship between different parameters, and effectively improve the accuracy and fault tolerance of perception; compared with the perception results of abnormal data input, the perception accuracy and recall rate are effectively improved; by training and testing the perception models based on time series parameters, image parameters, and time series and image parameter fusion, respectively, the perception results are similar or better than those of normal data input perception results.

[0028] The present invention adopts a power heterogeneous multi-parameter fusion model to perform data-level fusion of various power parameters, deeply fuses multiple parameters into power perception data, has a small amount of fusion information loss, high perception accuracy, and can extract information more accurately; at the same time, the fusion will better explore the relationship between meteorological conditions and thermal radiation, and further improve the reliability of early warning through the superposition and application of multiple parameters, thereby effectively improving the accuracy of pollution level discrimination, and the higher the degree of fusion, the more obvious the perception effect is. Based on the multi-parameter recursive graph, a graph convolutional neural network GCN is constructed to extract features and perform fusion; it is input into the judgment stage for final judgment. This method can avoid the poor detection effect at a certain stage and affect the final detection effect. It can effectively improve the accuracy and recall rate of insulator pollution level perception and the accuracy of perception when the humidity is high.

[0029] In order to overcome the limitations of traditional technology. The present invention uses a fire warning judgment module to segment the suspicious unknown areas of the image sequence where the fire occurred, and timely pre-judge and process them. After identification and confirmation, it determines whether the fire has reached the alarm level. Once the warning safety signal is detected, the multi-parameter warning judgment program will be automatically started, and the various information obtained in each stage will be comprehensively judged to give the final detection result. The background warning system will accelerate the issuance of the warning signal. This method has a fast convergence speed and can solve large-scale complex problems well. It is more efficient and accurate than traditional manual judgment. Compared with traditional methods, it has better detection results, higher detection rate and lower false detection rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0031] Figure 1 A schematic flow chart of a method for identifying fire in electric power equipment based on a multi-parameter recursive graph according to the present invention is shown. DETAILED DESCRIPTION

[0032] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0033] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] Now, exemplary embodiments according to the present application will be described in more detail with reference to the accompanying drawings. However, these exemplary embodiments may be implemented in a variety of different forms and should not be construed as being limited to the embodiments described herein. It should be understood that these embodiments are provided to make the disclosure of the present application thorough and complete, and to fully convey the concepts of these exemplary embodiments to those of ordinary skill in the art. In the accompanying drawings, for the sake of clarity, the thickness of the layers and regions may be enlarged, and the same reference numerals are used to represent the same devices, and thus their descriptions will be omitted.

[0036] Combination Figure 1 As shown, according to a specific embodiment of the present invention, a method for identifying fires in power equipment based on a multi-parameter recursive graph is provided.

[0037] Specifically, Figure 1 As shown, the power equipment fire identification method based on the multi-parameter recursive graph adopts the following steps:

[0038] S1: Use the intelligent fire detection algorithm module to obtain the electrical monitoring data of the target location and the target video image to be detected, wherein the target location includes the location of the monitoring area of ​​the power equipment fire warning intelligent monitoring system, and obtain the power image parameters based on the integration of the electrical monitoring data and the target video image. Based on the power image parameters, create an artificial intelligence-based fire recognition algorithm model.

[0039] S2: The fire identification algorithm model monitors video image segmentation and area detection, identifies the hidden danger electrical monitoring data of the target location, builds a deep neural network sub-model, and based on the deep neural network sub-model, performs feature extraction on the power image parameters to obtain target features, and creates a power heterogeneous multi-parameter fusion model based on the target features.

[0040] S3: Based on the power heterogeneous multi-parameter fusion model, the power image parameters and the power time series parameters are subjected to data-level fusion to obtain power perception data; based on the power heterogeneous multi-parameter fusion model, feature extraction is performed on the power perception data; based on the recursive algorithm module, the power time series parameters are arranged into a k×n recursive matrix, wherein k is the number of categories of the time series, and n is the length of the time series; based on the recursive matrix, a multi-parameter recursive graph is obtained; based on the multi-parameter recursive graph, a graph convolutional neural network is constructed to perform feature extraction.

[0041] S4: The fire warning judgment module is used to segment the suspicious unknown areas of the image sequence where the fire occurred for pre-judgment processing, and the judgment results of each feature are integrated. After identification and confirmation, it is determined whether the fire has reached the alarm level. When the warning safety signal is detected, the multi-parameter warning judgment program is automatically started, and the background warning system accelerates the issuance of the warning signal.

[0042] In an optional embodiment, the intelligent power fire monitoring and detection system consists of a main control computer and an electrical fire monitoring equipment video monitoring system. The main control computer collects telemetry and telesignal quantities in real time through a network security detection and early warning platform, centrally monitors and safely manages safety equipment, and grasps the operation of the system. The video monitoring system captures early warning accident information and realizes local real-time display of remote data. The data processed in the system is screened, combined and statistically generated into report data required by users, a database is established, and the electrical parameters collected as required are displayed to users in real time and accurately. At the same time, the collected data is stored in the database for users to query.

[0043] In another optional embodiment, the power equipment fire identification device includes: an intelligent fire detection algorithm module associated with a residual current electrical fire monitoring system, a recursive algorithm module, a fire warning judgment module, and a residual current fire alarm module formed by a residual current electrical fire monitoring detector, a temperature measurement electrical fire monitoring detector, and several bus isolators connected via a dual bus, wherein the residual current electrical fire monitoring system adopts a hierarchical distributed structure. When the ground fault current detected by the residual current fire alarm module in the protected line exceeds a preset value, after analysis and confirmation, an audible and visual alarm signal and a control signal are issued, and at the same time, the ground fault signal is transmitted to the main control machine through the bus within a few seconds. The main control machine sends out an audible and visual alarm signal and a control signal, the display screen displays the alarm address, and the alarm and control information is recorded and saved. The basic principle is that when the current, temperature and other parameters in the electrical equipment are abnormal or suddenly changed, the residual current electrical fire monitoring detector and the temperature measurement electrical fire monitoring detector terminal detection head use the electromagnetic field induction principle and the change of temperature effect to collect the information and transmit it to the detector. After amplification, conversion, analysis and judgment of the amplitude of the change, and comparison with the alarm setting value, once the setting value is exceeded, an alarm signal is issued and also transmitted to the equipment. After further identification and judgment by the equipment, when it is confirmed that a fire may occur, the host sends a fire alarm signal, lights up the alarm indicator, emits an alarm sound, and displays the fire alarm and other information on the LCD screen. The on-duty personnel will quickly go to the accident site for inspection and processing based on the above displayed information, and send the alarm information to the centralized control console.

[0044] In an optional embodiment, the intelligent fire detection algorithm module summarizes the detection objects according to the phenomenon of electrical fire, designs corresponding detection circuits to detect different detection objects, and selects a piezoelectric ceramic ultrasonic sensor with a center frequency of 40±1.0kHz, a direction angle of +60°, and a receiving sensitivity greater than or equal to -68 decibels to detect arcs. The center frequency of the ultrasonic sensor can meet the detection requirements of the system. For the residual current parameter, the residual current can be detected by the residual current sensor in the residual current type electrical fire monitoring detector. The residual current type electrical fire monitoring detector uses resistive current separation technology to accurately separate the resistive component leakage current representing the insulation performance from the residual current, filter out the capacitive component leakage current caused by the line and equipment, detect the pure resistive leakage current of the distribution line and implement the leakage alarm, so that the detection accuracy and alarm reliability are greatly improved.

[0045] Furthermore, the intelligent fire detection algorithm module is based on large-scale flame data recognition training and cooperates with cameras to identify fire conditions in the monitoring area in real time and capture early warning accident information. It realizes environmental monitoring by adopting video intelligent analysis and Internet of Things technology. On the basis of intelligent machine learning and processing of monitoring video images, preprocessing technology is used to eliminate interference factors in the image and highlight the main body of effective image information, laying the foundation for accurate detection and identification of fire image detection and identification of subsequent fire recognition algorithm models.

[0046] Furthermore, the intelligent fire detection algorithm module conducts targeted parameter detection and data collection on the area units covered by the target image. In the monitoring video image segmentation algorithm, the image is segmented into similar areas according to the standards for identification and detection based on the similarity and incoherence of the brightness values ​​of the image sample pixels. The maximum inter-class variance algorithm is used to calculate the segmentation threshold of the flame image and perform segmentation processing. For the joint image segmentation of the space domain and time domain, the Gaussian mixture model (GMM) is used to detect the flame target. The GMM applies a weighted combination of multiple unimodal distribution functions to approximate any probability distribution and estimate the parameters and weights of each mixed component. The flame binary image post-processing technology based on morphology is used to filter the image noise. Its purpose is to detect the flame target using the Gaussian mixture background model, improve the accuracy of the flame image segmentation, and transform the image information into a specific representation that is easy for the machine to understand and recognize after the flame image is efficiently segmented. This is to facilitate the subsequent intelligent fire detection and recognition processing.

[0047] Furthermore, the intelligent fire detection module combines the flame image segmentation results of the previous step, uses the marking algorithm and the external boundary tracking algorithm to measure the flame image, analyzes and calculates the brightness, contrast and smoothness of the grayscale image, and calculates the characterization evaluation element value of the texture feature based on the grayscale co-occurrence matrix.

[0048] Furthermore, the recursive algorithm module analyzes the periodicity, chaos and non-stationarity of the time series, reveals the internal structure of the time series, provides prior knowledge about the recursive state, information content and predictability of the system, and intuitively expresses the recursive phenomenon of the high-dimensional phase space trajectory with a recursive graph (RP). The recursive graph is obtained from a random system. The graphical characteristics of the recursive graph contain the development trend and law of the phase space trajectory over time. The recursive graph is expressed as a chaotic structure with inherent determinism. The key to constructing a recursive graph is to reconstruct the phase space. The so-called phase refers to the state of a system at a certain moment. All possible phase combinations of the system are collectively referred to as the phase space. The phase space reconstruction is to reconstruct the time domain information in the original phase space by applying relevant methods and changing relevant parameters. Through a series of transformations, the signal is promoted to a higher dimension, so that the signal produces certain system characteristics.

[0049] In the above embodiment, the recursion diagram is the recursion phenomenon of time j1 with respect to time j2 in the state space, which is usually represented by a two-dimensional time matrix. The traditional recursion matrix only has elements of 0 and 1, where 0 is white, indicating that the state difference between the two moments is large and does not constitute recursion; 1 is black, indicating that the state of the two moments is similar and constitutes recursion. The traditional method of constructing a recursion matrix is ​​for a single time series, and phase space reconstruction is required, and a suitable threshold needs to be selected to determine the elements of the recursion matrix.

[0050] Further, according to an embodiment of the present invention, the recursive algorithm module sets a recursive function to represent setting a graph of degree n starting at the position (x, y), and the starting positions are: (x, y+2 n-2 )、(x+2 n-2 , y) and (x+2 n-2 ,y+2 n-1 ), input a positive integer n, select the appropriate delay coefficient τ, embedding dimension m and threshold ε according to the phase space reconstruction, and iterate more than 100 times to obtain the x sequence as a recursive graph. The recursive graph is expressed by the following formula:

[0051] R ij =θ(ε-||x i -x j ||);

[0052] For the time series u k (k = 1, 2, ..., n), determine its sampling time interval as Δt, determine the appropriate embedding dimension m and delay time τ through relevant theoretical calculations, and then reconstruct the time series. The reconstructed dynamic system is:

[0053] x i =[u i ,u i +τ,…,u i +m-1)];

[0054] Calculate the i point x in the reconstructed phase space i and j point x j Distance:

[0055] S ij =||x i -x j ||;

[0056] Compute the recursive value:

[0057] R (i，j) =θ(ε i -S ij );

[0058] Among them, θ is the longest diagonal angle, i = 1, 2, ..., k, n-(m-1)τ, k is the time series category, n is the length of the time series, j = 1, 2, ..., n-(m-1)τ, and ||·|| represents the norm.

[0059] Furthermore, due to the different units and scales between the time series parameters, in order to synchronize the power time series parameters and image parameters in time, and considering the temporal and spatial characteristics of the power sensing target state, a multi-parameter recursive graph is generated using the time series. The specific process is as follows:

[0060] (1) Data matching: Assume that the power timing parameter is X i , let the nth sampling moment of the time series and the sampling moment of the image parameter be the same or close, that is, one image corresponds to a time series of length n, and introduce the quantitative recursive analysis method of recursive graph neural network (RecGNN) into the analysis of clutter signals, and apply it to obtain the k×n time series matrix of various recursive parameters;

[0061] (2) Normalization: When calculating the recursive matrix R nxn Before that, normalization is performed first. Specifically, recursive matrix normalization can be achieved by the following formula:

[0062] X i '=(X i -X i min) / (X i max- i Xmin);

[0063] Map the original data of the time series parameters of the above formula to the range between 0 and 1 to realize the recursive matrix R nxn ;

[0064] Among them, X i ' is the data in normalized form, X i is the time series parameter corresponding to the i-th type of parameter, X i min and X i max is the minimum and maximum value of the original data of the time series parameter, and k is the time series category;

[0065] (3) Calculate the multi-parameter recursive matrix: The k-type time series is regarded as k phase spaces. In order to avoid the selection of embedding dimension and delay time and further reduce the impact of parameter selection, the delay time is set to 0. At the same time, the norm of the difference between the states at two moments is used as the element of the recursive matrix to further reduce the impact of parameter selection.

[0066] (4) Generate a recursive graph: Use the elements in the multi-parameter recursive matrix as the pixel values ​​of the image to generate a multi-parameter recursive graph, use different convolutional neural networks to extract features from the recursive graph and the image, construct a multi-parameter recursive graph feature extraction network, and then use different convolutional neural networks to extract features from the recursive graph and the image respectively;

[0067] (5) Call the function itself in the recursive function. Each recursive call creates a new function instance. Connect these function instances in a certain order to form a recursive chain. Based on the feature-level fusion model of power time series parameters and image parameters, convert the power time series parameters into a multi-parameter recursive graph. After reconstructing the phase space of the time series, select different neighborhood recursive chain radii to obtain different multi-parameter recursive graphs. Based on the weight, the two types of features are fused and the target perception, transmission line ice level perception and insulator contamination level perception are performed to obtain one or more fire warning parameters at the current time of the warning point. Based on the recursive neural network multivariate time series prediction, the transmission line ice level perception and insulator contamination level perception are used as application scenarios, and the warning signal is output to the display module in multiple graphs and multiple indicators.

[0068] The fire recognition algorithm model is based on the artificial intelligence perception information processing algorithm and adopts the deep learning method. The fire smoke detection algorithm based on deep learning is used to train and identify the fire picture set collected by the intelligent fire detection module, and a large number of image learning classifiers are trained; the different features of the image are identified through the convolution layer, the features of the image are extracted, and the feature representation and relationship of the image are learned and learned and trained according to the fuzzy neural network algorithm to complete the learning of fire features and automatically generate membership function parameters suitable for the fire detection model; after the training, the weights of the relative effects of each fuzzy inference rule are extracted, and the mapping rules between the fire parameters and the sensor input mode are found through the change of weights, the images are classified, and the current state is associated with the previous state using a recurrent unit, each element in the sequence is processed, the current element is predicted, the next element of a sequence is predicted, and the next value of a time series is predicted, and a deep neural network sub-model is generated according to the training data.

[0069] The above purpose is to combine fuzzy system and neural network according to the characteristics of fire detection, realize automatic adjustment of fuzzy system design parameters, and apply the fuzzy neural network algorithm to fire detection to reduce the false alarm rate of fire.

[0070] In an optional embodiment, the fire identification algorithm model is based on the flame detection algorithm of wavelet transform and fuzzy support vector machine (FSVM), performs two-dimensional wavelet transform on the image, extracts the square sum of wavelet coefficients of several frames of pixel points as eigenvalues, and applies fuzzy support vector machine as classifier to determine whether it is a pixel point in the flame area. If it is, the support vector machine is trained using the above-mentioned extracted fusion features, and the extracted motion area is sequentially classified into smoke and non-smoke through the classifier. After detecting the flame or smoke, the detected position information and category information are sent to the deep neural network sub-model, and the fuzzy neural network is established with the smoke concentration and smoke concentration change rate signal as input, and finally connected to the classification or regression layer for target perception. This method has strong robustness to noise, interference, etc., and still has good classification effect when the training data contains outliers.

[0071] At present, most video acquisition devices capture images based on RGB color space, but the components of this space are highly correlated, have much redundant information, and require a lot of calculation. Considering that the conversion process from RGB space to YUV space is linear, and the brightness signal Y and chrominance signals U and V in YUV space are independent of each other.

[0072] Therefore, in another optional embodiment, the fire recognition algorithm model first extracts flame features in the transform domain, calculates the time series of the farthest distance of the edge of the candidate flame area, and performs a two-dimensional Fourier transform on the sequence to obtain the frequency change characteristics, and inputs the R value of each candidate pixel in the RGB color space or the Y value (grayscale value) in the YUV color space into the corresponding level filter for time wavelet analysis, and selects to combine morphological operations in the YUV color space to obtain a shadow pixel point set and a foreground target pixel point set with high confidence, and the number of zero crossing points of the output wavelet sub-signal is used to distinguish whether it is a flame pixel point. At the same time, the mean of the spatial wavelet transform is compared with a fixed threshold, and the comparison result is also used as one of the bases for judging whether a flame exists.

[0073] Furthermore, each neuron of the deep neural network sub-model is connected to the input signal of each neuron in the previous layer. In each connection process, the signal from the previous layer is multiplied by a weight and a bias is added. Then, through a nonlinear activation function and multiple combinations of simple nonlinear functions, all feature mappings of different layers of the convolutional neural network from the input space to the output space are realized. Then, the deep model is trained after calibration with smoke and flames in the fire sample image. Then, the convolutional neural network CNN is used to extract features of the two types of parameters, namely, time series parameters and image parameters. The feature matrices of the time series parameters and image parameters are concatenated and fused according to the weights to form a fused feature matrix. Then, the matrix is ​​used as input and passes through the fully connected layer. The weight factors W and W2 are used as training parameters. For the regression problem, in order to avoid interference from human factors, the weight factors are used as network parameters, and iterative training is performed with the cross entropy loss or the minimum mean square error target until convergence.

[0074] In the scenario of intelligent fire identification and monitoring, the fire warning judgment model is based on a multi-feature fusion smoke detection algorithm. It uses an artificial neural network as a fire identification algorithm model that can simulate the brain's thinking mode for intelligent judgment. It takes the characterization parameters of the flame image segmentation area as input values, and uses the artificial neural network algorithm model to quickly eliminate environmental interference factors. It automatically adjusts the operating parameters according to on-site changes, and accurately detects, identifies and judges fires with image recognition speed and accuracy close to the human brain's thinking level. It also compares and analyzes with fire occurrence standards to obtain comparison results, determine the actual fire point, and issue early warnings in a timely manner.

[0075] Furthermore, in the fire warning judgment module of the monitored scene, the brightness algorithm is first used to calculate the brightness of the original image without flames, and it is set as the brightness threshold. The set brightness value is compared with the brightness value calculated by the image of other scenes. If it is greater than the brightness threshold, there may be a fire; if it is less than the brightness threshold, there is no fire, thereby judging whether there is a fire in the detection scene. Set the detection threshold. When processing smoke images and flame images, the collected images are grayed, and after the pixel value analysis and processing, if the average brightness value obtained is higher than the detection threshold, it can be judged that there is a fire in the image, and an alarm window pops up for prompting; if it is lower than the detection threshold, there is no fire in the image. For example, in the monitored scene, when there is no fire, the original brightness value is calculated according to the function of calculating the average brightness in image processing, and this value is set as the detection threshold. The collected images are grayed, and after the pixel value analysis and processing, if the average brightness value obtained is higher than the detection threshold, it can be judged that there is a fire in the image, and an alarm window pops up for prompting; if it is lower than the detection threshold, there is no fire in the image.

[0076] According to another aspect of the present invention, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the above method is executed when the program is run.

[0077] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium including a stored executable program, wherein when the executable program is running, the device where the storage medium is located is controlled to execute the above method.

[0078] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, and the computer program implements the above method when executed by a processor.

[0079] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0080] In addition to the above, it should be noted that "one embodiment", "another embodiment", "embodiment", etc. mentioned in this specification refer to the specific features, structures or characteristics described in conjunction with the embodiment included in at least one embodiment generally described in this application. The same expression appearing in multiple places in the specification does not necessarily refer to the same embodiment. Further, when describing a specific feature, structure or characteristic in conjunction with any embodiment, it is claimed that the realization of such feature, structure or characteristic in conjunction with other embodiments also falls within the scope of the present invention.

[0081] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying fire in power equipment based on multi-parameter recursive graph, characterized in that: include: An intelligent fire detection algorithm module is used to obtain electrical monitoring data of a target location and a target video image to be detected, wherein the target location includes the location of a monitoring area of ​​an intelligent monitoring system for early warning of fire of electric power equipment. Electric power image parameters are obtained based on the integration of the electrical monitoring data and the target video image. The electrical monitoring data includes current and temperature parameters in the electric equipment. Based on the electric power image parameters, a fire recognition algorithm model is created, and image segmentation and area detection are performed on the target video image to obtain a fire image of a suspicious area. Based on the suspected area fire image, the marking algorithm and the external boundary tracking algorithm, the hidden danger electrical monitoring data of the target position in the suspected area fire image is measured, and the first target feature of the suspected area fire image is determined based on the measurement result; based on the first target feature, a deep neural network sub-model is built; Based on the deep neural network sub-model, feature extraction is performed on the power image parameter to obtain a second target feature, based on the second target feature, a power heterogeneous multi-parameter fusion model is created, based on the power heterogeneous multi-parameter fusion model, data-level fusion is performed on the power image parameter and the power timing parameter to obtain power perception data, and feature extraction is performed on the power perception data to obtain a third target feature; Based on the recursive algorithm module, the power timing parameters are arranged as k×n The recursive matrix of , where k is the number of categories of the time series, n is the length of the time series, obtaining a multi-parameter recursive graph based on the recursive matrix, constructing a graph convolutional neural network based on the multi-parameter recursive graph, performing feature extraction, and obtaining a fourth target feature; A fire warning judgment module is used to segment the suspicious unknown area where the fire occurs in the image sequence, and the second target feature, the third target feature, and the fourth target feature are judged to obtain a pre-judgment processing result, wherein the judgment process includes: judging the second target feature, the third target feature, and the fourth target feature based on a smoke detection algorithm with multi-feature fusion to obtain a brightness value and an average brightness value; obtaining a pre-judgment processing result according to the brightness value, the brightness threshold, the brightness threshold and the detection threshold; obtaining an alarm result according to the pre-judgment processing result; according to the alarm result, the control system starts a warning program and outputs a warning signal.

2. The method for identifying fire in electric power equipment based on multi-parameter recursive graph according to claim 1, characterized in that: The electric equipment fire identification method adopts an electric equipment fire identification device to identify fire, and the electric equipment fire identification device includes: an intelligent fire detection algorithm module associated with a residual current electrical fire monitoring system, a recursive algorithm module, a fire warning judgment module, and a residual current fire alarm module formed by a residual current electrical fire monitoring detector, a temperature measurement electrical fire monitoring detector, and multiple bus isolators connected via a dual bus, wherein the residual current electrical fire monitoring system adopts a hierarchical distributed structure, when the ground fault current detected by the residual current fire alarm module in the protected line exceeds a preset value, after analysis and confirmation, an alarm signal and a control signal are issued, and at the same time, the ground fault signal is transmitted to a main control machine through the bus within a few seconds, the main control machine sends an alarm signal and a control signal, the display screen displays the alarm address, and the alarm and control information are recorded and saved.

3. The method for identifying fire in electric power equipment based on multi-parameter recursive graph according to claim 1, characterized in that: The intelligent fire detection algorithm module is obtained based on flame data recognition training. The intelligent fire detection algorithm module cooperates with a camera to obtain the target video image of the target position in real time, and uses preprocessing technology to eliminate interference factors in the target video image to highlight the effective information body of the target video image.

4. The method for identifying fire in electric power equipment based on multi-parameter recursive graph according to claim 3 is characterized in that: Based on the intelligent fire detection algorithm module, the parameter data of the area unit covered by the target video image is obtained, the target video image is segmented into similar areas based on the video image segmentation algorithm and identification and detection are performed, and the segmentation threshold of the flame image is calculated based on the maximum inter-class variance algorithm and segmentation processing is performed; for the joint image segmentation of the space domain and time domain, the Gaussian mixture model is used to detect the flame target, the Gaussian mixture model applies a weighted combination of multiple unimodal distribution functions to approximate any probability distribution, estimates the parameters and weights of each mixed component, and uses the morphology-based flame binary image post-processing technology to filter image noise.

5. The method for identifying fire in electric power equipment based on multi-parameter recursive graph according to claim 1, characterized in that: Based on the fire picture set collected by the intelligent fire detection module, the fire recognition algorithm model is trained using an artificial intelligence perception information processing algorithm and a deep learning fire smoke detection algorithm; based on the fire recognition algorithm model, a fuzzy neural network method is used to extract the features of the fire picture set, automatically generate membership function parameters and weights of the relative effects of fuzzy inference rules, and obtain a mapping rule between the features of the fire picture set and fire parameters; and according to the mapping rule, the target video image is classified into a smoke image and a non-smoke image.

6. The method for identifying fire in electric power equipment based on multi-parameter recursive graph according to claim 5, characterized in that: The fire identification algorithm model is based on the flame detection algorithm of wavelet transform and fuzzy support vector machine (FSVM). It performs two-dimensional wavelet transform on the target video image, extracts the square sum of wavelet coefficients of several frames of pixel points as the eigenvalue, and applies fuzzy support vector machine as a classifier to determine whether the extracted pixel points are flame area pixels. If they are, the extracted fusion features are used to train the support vector machine, and the extracted motion areas are classified into smoke and non-smoke through the classifier in turn. After detecting flame or smoke, the detected position information and category information are sent to the deep neural network sub-model, and the fuzzy neural network is established with smoke concentration and smoke concentration change rate signals as input, and finally connected to the classification or regression layer for target perception.

7. The method for identifying fire in electric power equipment based on multi-parameter recursive graph according to claim 1, characterized in that: The fire recognition algorithm model first extracts flame features in the transform domain, calculates the time series of the farthest distance of the edge of the candidate flame area, and performs a two-dimensional Fourier transform on the series to obtain the characteristics of frequency change. The R value of each candidate pixel in the RGB color space or the Y value in the YUV color space is input into the corresponding level filter for time wavelet analysis, and the shadow pixel point set and the foreground target pixel point set are obtained by combining morphological operations on the YUV color space. The number of zero crossing points of the output wavelet sub-signal is used to distinguish whether it is a flame pixel point. At the same time, the mean value of the spatial wavelet transform is compared with a fixed threshold, and the comparison result is also used as one of the bases for judging whether a flame exists.

8. The method for identifying fire in electric power equipment based on multi-parameter recursive graph according to claim 1, characterized in that: In the scenario of intelligent fire identification and monitoring, the fire warning judgment module is based on a multi-feature fusion smoke detection algorithm, and uses an artificial neural network as a fire identification algorithm model that can simulate the brain's thinking mode for intelligent judgment. The characterization parameters of the flame image segmentation area are used as input values, and the artificial neural network algorithm model is used to quickly eliminate environmental interference factors, automatically adjust operating parameters according to on-site changes, detect, identify and judge fires, and compare and analyze them with fire occurrence standards to obtain comparison results, determine the situation of the actual fire point based on the comparison results, and issue warnings in time.

9. The method for identifying fire in electric power equipment based on multi-parameter recursive graph according to claim 1, characterized in that: In the monitored scene, the fire warning judgment module first uses the brightness algorithm to calculate the brightness of the original image without flames, and sets the brightness of the original image without flames as the brightness threshold. The brightness threshold is set and compared with the brightness value calculated by the image of other scenes. If the brightness value calculated by the image of other scenes is greater than the brightness threshold, a fire may occur; if the brightness value calculated by the image of other scenes is less than the brightness threshold, no fire occurs. When processing smoke images and flame images, the collected images are grayed. After pixel value analysis and processing, if the average brightness value obtained is higher than the detection threshold, it can be judged that there is a fire in the image, and an alarm window pops up for prompting; If the value is below the detection threshold, there is no fire in the image.

10. An electronic device, characterized in that: include: A memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 9 when running.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is executed, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 9.

12. A computer program product, characterized in that The invention comprises a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 9.

Images