Method and system for identifying fire combustion form of oil-filled equipment of transformer substation

Abstract

The invention relates to a method and system for identifying a fire combustion form of oil-filled equipment of a transformer substation, and the method comprises the steps: carrying out the binarization processing of an original image, and segmenting a combustion region; calculating the area, height, width, centroid, number of connected regions and other parameters of the segmented region image; and inputting the parameters into a pre-established and trained improved particle swarm algorithm-support vector machine model, and finally obtaining a classification identification result of the fire combustion form of the oil-filled equipment of the transformer substation. According to the improved particle swarm algorithm-support vector machine model, an evolution rule of a genetic algorithm is added in a traditional particle swarm algorithm, the ability of the particle swarm algorithm in the aspect of optimization is optimized, the probability of falling into a local optimal solution is low, iterative search optimization is conducted on support vector machine kernel parameters, and the optimal kernel parameters are obtained. The method can identify the combustion form in time at the early stage of fire combustion of the oil-filled equipment of the transformer substation, and assists fire-fighting departments and intelligent fire-fighting equipment to carry out targeted fire extinguishing operation.

Description

Technical Field

[0001] The invention relates to the field of fire safety of transformer substations, and in particular to a method and system for identifying the burning form of fire in oil-filled equipment of transformer substations. Background Art

[0002] The fire and explosion accident of oil-filled equipment in the substation is the most serious power disaster. Once a fire or even an explosion occurs, it often causes casualties and huge economic losses, and also seriously affects the operation of the power grid. According to statistics, the probability of transformer fire in my country in the past 20 years is about 0.01%-0.03% / year, and the probability of transformer fire in foreign countries is 0.04%-0.25% / year. With the implementation of networking, integration and intelligent technical transformation of substation monitoring systems in my country, and the promotion and use of various fire prevention and control technologies, substation fire and explosion accidents are on a downward trend, but on the whole, substation fire and explosion accidents continue to occur, indicating that my country's substation fire prevention and control is still facing a severe situation, and major accidents have not been effectively curbed. Transformer fires have the phenomenon of coexistence of multiple combustion forms such as oil pool fire, flowing fire and jet fire. There are great differences in the combustion characteristics of different combustion forms, which brings great challenges to firefighting. Therefore, intelligent identification of the combustion form of transformer fire is of great significance for timely and accurate formulation of measures during the fire period.

[0003] Prior art document 1 (CN109711345A, a flame image recognition method, device and storage medium thereof) provides a flame image recognition method, including: determining a suspected flame area in a captured image; using a filter to extract a texture feature image of the suspected flame area; extracting local binary features of the texture feature image based on a local binary pattern; using a support vector machine (SVM) classifier to determine whether there is a flame in the suspected flame area based on the local binary features; the main purpose is to identify whether a fire has occurred in the area to be tested, but it cannot judge the form of the fire, and cannot assist firefighters in formulating a complete and targeted fire extinguishing plan.

[0004] The present invention specifically studies the characteristic value differences reflected by the fires of oil-filled equipment of different scales in substations, analyzes the intra-class / inter-class characteristic relationships of the fire forms of substations, establishes a multi-source heterogeneous database of oil-filled equipment fires in substations, and uses it as the driving source to extract the spatial domain feature information and time-frequency domain feature information of the substation oil-filled equipment fire image based on the convolutional neural network DeeplabV3+-MobilenetV2 model, so as to perform more accurate flame combustion form recognition. The original feature extraction network of DeeplabV3+ is Xception, but after comparative testing, it is found that when MobilenetV2 is used as its feature extraction network, on the basis of the small difference in parameters such as IOU and Recall, its parameter quantity is reduced to 1 / 10 of the original, and FPS is also greatly improved, which provides a better choice for flame segmentation and recognition. The present invention is a targeted study based on the oil pool fire, flowing fire, and jet fire in the development process of substation fire. In the combustion process of transformer fire: in the primary stage of fire, transformer fire is mainly oil pool fire. By establishing a database, image processing and other means, the height and width of the flame are determined for identification. In the stable stage of fire, transformer fire forms a phenomenon of coexistence of multiple combustion forms, including oil pool fire, flowing fire and initial jet fire. In the sudden change stage of fire combustion, the box body ruptures, the jet fire phenomenon is prominent, and the thickness of the flowing fire flame on the side wall increases significantly. Flowing fire and jet fire generally appear simultaneously, mostly on the upper surface and side walls. When a jet fire occurs in a transformer fire, the internal pressure of the box body is generally large, and the flame height will increase accordingly. In addition, the flame height of the transformer fire is closely related to the internal pressure and is mutational. Therefore, studying the relationship between flame height and pressure changes is convenient for identifying jet fires. The present invention mainly studies the unique flame combustion forms of substation fires by establishing training sets, processing images, calculating parameters, etc.

[0005] Prior art document 2 (CN 116935279A, application publication date 2023.10.24) discloses a method for identifying the development stage of a substation fire based on dynamic flame texture, including establishing a linear dynamic system model of a substation oil-filled equipment fire to characterize the dynamic characteristics of the texture and designing a kernel function based on the main angle of the subspace to measure the similarity of the flame dynamic model. The present invention regards the time-varying flame texture of the oil-filled equipment fire as a dynamic system, couples the temporal dynamic features of the image sequence with the image features, and determines the development stage of the oil-filled equipment fire by establishing a generative model based on dynamic flame texture. Summary of the invention

[0006] In order to solve the deficiencies in the prior art, the present invention provides a method and system for identifying the combustion form of oil-filled equipment fire in a substation, so as to overcome the defects in the prior art that the combustion form cannot be identified, resulting in inefficient and blind fire extinguishing.

[0007] The present invention adopts the following technical solution.

[0008] The present invention provides a method for identifying the combustion form of oil-filled equipment fire in a substation, comprising:

[0009] Image acquisition: Collect surveillance camera videos and experimental images of individual combustion during the substation fire, and use video framing technology to extract the flame image per second in the video;

[0010] Flame area segmentation: The images obtained by dividing the above video frames are binarized through Python programming. Based on the image binarization technology, an adaptive threshold function is introduced to process the original image of the data set to obtain the corresponding flame burning area;

[0011] Flame feature extraction: Based on the connected region detection method, the area, height, width, centroid and number of connected regions of the binary image of the flame combustion region are calculated;

[0012] Acquisition of other fire parameters: Collect other parameters that characterize the combustion form of oil-filled equipment fires, mainly including temperature, smoke composition and internal pressure of the equipment;

[0013] Dataset construction steps: The area, height, width, centroid and number of connected regions of the binary image of the flame burning area at the same time are combined with other fire parameters, and the combustion form is marked. The oil pool fire is marked as 1, the flowing fire is marked as 2, and the jet fire is marked as 3. Finally, it is divided into training set and test set in a ratio of 7:3;

[0014] Combustion form identification: the data of the training set samples are input into a pre-established and trained improved particle swarm algorithm-support vector machine model to predict the classification labels and obtain the classification identification results of the flame combustion form.

[0015] Preferably, the connected regions are regions with the same pixel values ​​and adjacent to each other. The number of connected regions is used to describe the splitting of the flame. A plurality of connected regions indicates that the flame has a plurality of burning points or the shape of the flame is irregular and the flame burns in a dispersed manner.

[0016] Preferably, the other flame parameter extraction includes:

[0017] The temperature includes oil temperature, external axial temperature of the oil-filled equipment and side wall temperature, which can be obtained by connecting a data acquisition instrument with a K-type thermocouple, and directly obtaining the temperature and heat flux density data signals as one-dimensional data; the internal pressure of the oil-filled equipment can be obtained through a pressure transmitter; the changes in the flue gas composition can be obtained through a flue gas collector to monitor the changes in the CO and CO2 content.

[0018] Preferably, the combustion form identification is based on an adaptive weighted particle swarm algorithm and a GA-PSO algorithm is introduced. In each iteration, the fitness values ​​of the particles are sorted from large to small. The larger the fitness value of the particle, the better its adaptability. The fitness function is the average value of the classification accuracy under 3-fold cross validation. First, particles with higher fitness values ​​are selected to directly enter the next generation to ensure that the next generation of particles is relatively better. At the same time, the positions and velocities of the first half of the particles with higher fitness values ​​are assigned to the second half of the particles with lower fitness values. These particles are randomly crossed in pairs, and then the same number of offspring particles are generated, and the offspring particles replace the parent particles.

[0019] Preferably, the position and velocity of the offspring particles are calculated as follows:

[0020]

[0021] Where p is a random number between 0 and 1, child(x) represents the position of the child particle, child(v) represents the velocity of the child particle, parent1(x), parent2(x), parent1(v), parent2(v) represent the position and velocity of the parent particle respectively.

[0022] Preferably, the SVM algorithm is introduced into the combustion form recognition process to carry out specific training of the model, including:

[0023] The samples are trained using the libsvmtrain function of the LIBSVM toolbox, model = libsvmtrain(trainlabel, trainfinal, ['options']), where model is the obtained training model; trainlabel is the known category label, which is an N*1 matrix; trainfinal is the input sample data, which is the area, height, width, centroid and number of connected regions of the binary image obtained by flame segmentation and feature extraction, combined with other fire parameters, and is an N*M numerical matrix, where N is the number of sample points and M is the feature dimension of the sample; ['options'] is the optional operation parameter of the libsvmtrain function, including the category of the kernel function, the penalty parameter c and the kernel parameter g value. The selected classification kernel function is the RBF kernel function, and the parameters c and g are determined by the optimization function.

[0024] Preferably, the working process of the flame combustion form recognition model includes:

[0025] (1) In the input training set sample data, the speed, position, and all related parameters of the population particles are randomly initialized; the initial fitness values ​​of all particles are calculated so that the particles can find the optimal solution for the flame combustion form identification parameters;

[0026] (2) Find the initial extreme value, determine the individual extreme value pbest and the group extreme value gbest according to the fitness value of the initial particles, prevent only finding the local optimal solution of the flame combustion form identification parameters, and facilitate the particles to gather roughly at the optimal solution;

[0027] (3) Update the speed and position of the offspring particles according to the proposed crossover formula and replace the parent particle position to further capture the flame combustion data;

[0028] (5) Perform adaptive particle mutation;

[0029] (6) Update inertia weight;

[0030] (7) If the stopping condition is met or the maximum number of iterations is exceeded, the optimal solution is output; otherwise, jump to step (3) to continue searching.

[0031] The present invention also provides a system for identifying the combustion form of oil-filled equipment fire in a substation, which is used in the aforementioned method for identifying the combustion form of oil-filled equipment fire in a substation, and is characterized by comprising:

[0032] Image acquisition module, used to collect surveillance camera videos during the substation fire and experimental images of separate combustion forms;

[0033] Flame area segmentation module, used to obtain the flame combustion area;

[0034] A flame feature extraction module is used to extract image features of the flame combustion area;

[0035] The combustion form recognition module is used to classify and recognize the flame combustion form.

[0036] The present invention also provides a terminal, including a processor and a storage medium;

[0037] The storage medium is used to store instructions;

[0038] The processor is configured to operate according to the instructions to execute the steps according to the aforementioned method.

[0039] The present invention also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

[0040] The beneficial effects of the present invention are that, compared with the prior art, the present invention has the following advantages:

[0041] (1) The method for identifying the combustion form of oil-filled fire equipment in a substation of the present invention can timely identify the combustion form of the substation fire in its early stage through the improved particle swarm-support vector machine model, and generate the fire burning area, height, width and combustion form in real time. The algorithm has high optimization efficiency and strong recognition performance, and can effectively identify and classify the combustion form of flames, and assist firefighters in fire departments to carry out targeted firefighting operations.

[0042] (2) The structure of the particle swarm algorithm (PSO) is relatively simple, easy to operate and implement, with a fast convergence speed, and no complex operations such as selection, crossover and mutation. However, the particle swarm algorithm has the problems of easy premature convergence, relatively low classification accuracy, and easy to fall into the local optimal solution. In order to avoid the above problems in the identification of the combustion form of substation fire, the present invention combines the advantages of the PSO and GA algorithms, and proposes a hybrid optimization algorithm (GA-PSO) based on the adaptive weighted particle swarm algorithm and introduces the crossover and mutation operator of the genetic algorithm (GA) to improve the search performance of the particle swarm algorithm. The GA-PSO algorithm maximizes the advantages of the two algorithms and avoids the shortcomings of the particle swarm algorithm. The algorithm increases the diversity of the population and increases the original search range of the particle swarm algorithm, improving the defects of the particle swarm algorithm in the search process, such as premature convergence and easy to fall into the local optimal value.

[0043] (3) The method for identifying the combustion form of oil-filled equipment fire in a substation of the present invention can timely identify the combustion form of the substation fire in its early stage through the improved particle swarm-support vector machine model, and generate the fire burning area, height, width and combustion form in real time. The algorithm has high optimization efficiency and strong recognition performance, and can effectively identify and classify the combustion form of flames, and assist firefighters in fire departments to carry out targeted firefighting operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of a method for identifying the combustion form of a fire in oil-filled equipment in a substation according to the present invention;

[0045] Figure 2 It is a schematic diagram of the GA-PSO algorithm flow in the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0047] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0048] Example 1

[0049] The present invention provides a method for identifying the combustion form of oil-filled equipment fire in a transformer substation, comprising the following steps:

[0050] Step 1, image acquisition: collect surveillance camera videos and experimental images of individual combustion during the substation fire, and use video frame segmentation technology to extract the flame image per second in the video;

[0051] Step 2, flame area segmentation: binarize the image obtained by dividing the video into frames through Python programming, introduce an adaptive threshold function based on the image binarization technology, process the original image of the data set, and obtain the corresponding flame burning area;

[0052] Step 3, flame feature extraction: Based on the connected region detection method, the area, height, width, centroid and number of connected regions of the binary image of the flame combustion region are calculated.

[0053] The connected regions are regions with the same pixel values ​​and adjacent to each other. The number of connected regions is used to describe the splitting of the flame. A plurality of connected regions indicates that the flame has a plurality of burning points or the shape of the flame is irregular and the flame burns in a dispersed manner.

[0054] By extracting the flame feature information in the binary image and calculating the pixel area occupied by the flame, the two-dimensional image information is converted into a one-dimensional flame pixel area. Python programming is used to traverse the white area in the binary image and calculate the area occupied by the white pixels.

[0055] In order to calculate the centroid position of the flame area, the moment of the image is needed. The moment is a digital feature used to describe a random variable. If the pixel coordinates of the image are regarded as two-dimensional random variables (X, Y), the grayscale image can be represented by a two-dimensional grayscale density function, and its characteristics can be described by the moment. In OpenCV, the formula for the centroid coordinates is:

[0056] M il =Σx,y(array(x,y)×x i ×y i )

[0057] The coordinates of the center of mass of the figure are , i=0,1,2,3,…,represents the number of columns, l=0,1,2,3,…,represents the number of rows, where, m01 and m 10 is the first-order moment, m 00 is the zero-order moment. The moments() method in OpenCV can be used to obtain the center distances of each order of a known contour. This method and formula can be used to obtain the center of mass position of the flame.

[0058] Step 4, acquisition of other fire parameters: collect other parameters that characterize the combustion form of the oil-filled equipment fire, mainly including temperature, smoke composition and internal pressure of the equipment; among them, temperature includes oil temperature, external axial temperature of the oil-filled equipment and side wall temperature, which can be obtained by connecting a data acquisition instrument with a K-type thermocouple, and directly obtaining temperature and heat flux density data signals as one-dimensional data; the internal pressure of the oil-filled equipment can be obtained through a pressure transmitter; changes in smoke composition can be obtained through a smoke collector, which mainly monitors changes in CO and CO2 content.

[0059] It should be noted that the three main combustion forms of transformer oil-filled equipment fires, namely oil pool fire, jet fire and flowing fire, have great differences in image, temperature and pressure. Before the jet fire occurs, the pressure in the oil-filled equipment will rise rapidly, causing the equipment to rupture and gap, and then a jet fire will occur. The axial temperature of the oil-filled equipment will change greatly, and the flame will be "inverted triangle" or "cylindrical"; flowing fire is a fire form in which the oil flows and burns along the side wall of the oil-filled equipment. The temperature of the side wall will change greatly, and the flame will spread downward in a rectangular shape. Oil pool fire generally occurs in the middle of the oil-filled equipment fire. The leaked transformer oil gathers and ignites at the bottom of the oil-filled equipment to form an oil pool. The flame shape of the oil pool fire is relatively stable, and it is obviously different from other combustion forms in the image. Therefore, for the identification of the fire form of transformer oil-filled equipment, accurate distinction can be achieved through four convenient parameters: image, temperature, pressure and smoke.

[0060] Among them, image recognition of fire source is the most important recognition method. Use infrared camera to collect images, perform binary processing on the original images, segment the burning area to identify the flame. The three convenient parameters of temperature, pressure and smoke are only used as auxiliary judgment. Temperature is collected from the oil pool in the oil collection pit, the wall / top surface / top opening of the converter body, the temperature field changes of the wall of the firewall, and the inside of the converter; smoke is collected through the changes in smoke and oxygen concentration during the combustion and fire extinguishing process; by installing a pressure sensor in the experiment, the pressure changes inside the box during the development of the fire are measured;

[0061] For oil pool fire, the basic combustion characteristics test of transformer oil fire was carried out to analyze the transformer oil pool fire plume structure, plume temperature and flame pulsation frequency. The combustion development stages of transformer oil pool fire (i.e. ignition, growth, full combustion and extinction) and the spatial structure of fire plume (continuous flame area, intermittent flame area and buoyancy plume area) were obtained, the periodic pulsation law of transformer oil pool fire flame was revealed, and it was determined that the pulsation frequency was inversely proportional to the oil pool diameter, which can better distinguish the combustion of oil pool fire.

[0062] For flowing fire, a transformer oil flowing fire combustion test bench was established to study the occurrence conditions, development process and characteristic parameters of flowing fire. The parameters such as transformer oil temperature, wall temperature, fire plume temperature, and thermal radiation intensity were obtained during its development. Through the coupling relationship between the initial oil temperature and oil flow rate variables during the development of flowing fire, the critical flow rate of transformer oil in the early stage of flowing fire formation under the current test conditions was clarified, and the critical flow rate was related to the initial oil temperature and wall temperature, which can better distinguish flowing fire.

[0063] For jet fire, a jet fire combustion experiment of transformer oil in oil-filled equipment of substation was carried out to study the formation mechanism and combustion characteristics of jet fire (spray fire) of oil-filled equipment transformer oil under the action of external heat source. By establishing the dimensionless flame height and dimensionless fire source power function relationship: L / D = 1.8Q^2 / 5-1.3, analyzing the axial centerline temperature data of the fire plume and obtaining the axial centerline temperature distribution function relationship of the fire plume: T = -1384.74p2+1515.71P+c, a prediction model for the radiation heat flux of jet fire of oil-filled equipment in substation was established, which played a better role in the identification of jet fire.

[0064] Step 5, data set construction step: The area, height, width, centroid and number of connected regions of the binary image of the flame burning area at the same time are combined with other fire parameters, and the burning form is marked. Specifically, the oil pool fire is marked as 1, the flowing fire is marked as 2, and the jet fire is marked as 3. Finally, it is divided into training set and test set in a ratio of 7:3.

[0065] Step 6, combustion form identification: input the data of the training set samples into the pre-established and trained improved particle swarm algorithm-support vector machine model to predict the classification label and obtain the classification identification result of the flame combustion form.

[0066] Among them, PSO determines the search based on its own speed. It retains the global search method based on population particles and uses a speed-displacement model. The algorithm is easy to implement and has few parameters, but there are still omissions caused by simplicity and it is not precise enough. Therefore, the present invention improves it, adds a crossover strategy based on fitness value selection, enhances the global search ability of particles, and adds adaptive particle mutation, which theoretically strengthens the search ability of particles in the later stage.

[0067] like Figure 2 As shown, the present invention proposes a hybrid optimization algorithm (GA-PSO) based on an adaptive weighted particle swarm algorithm and introducing a genetic algorithm (GA) crossover mutation operator to improve the search performance of the particle swarm algorithm. The basic idea is that the particle swarm algorithm is used to mainly explore the search space in the early stage in order to reach the optimal area as soon as possible, and a crossover strategy based on fitness value selection is incorporated to improve the particle swarm algorithm and enhance the global search ability of particles. In the later stage, the model is improved by adding an adaptive particle mutation implementation method, mainly developing the optimal area in order to find the optimal solution as soon as possible. The GA-PSO algorithm is used to optimize the model as a whole.

[0068] The crossover strategy based on fitness value selection simulates the phenomenon that the offspring population inherits the excellent characteristics of the parent generation in natural evolution, and strengthens the diversity of the population. Therefore, the introduction of the crossover operation theoretically strengthens the global search ability of the particles.

[0069] In the new algorithm constructed by the present invention, the specific idea of ​​crossover is: in each iteration, the fitness values ​​of the particles are sorted from large to small. The larger the fitness value of the particle, the better its adaptability. Its fitness function is the average value of the classification accuracy under 3-fold cross validation. First, the particles with higher fitness values ​​are selected to directly enter the next generation to ensure that the next generation of particles is relatively better; at the same time, the position and speed of the first half of the particles with higher fitness values ​​are assigned to the second half of the particles with lower fitness values. These particles are randomly crossed in pairs, and then the same number of offspring particles are generated, and the offspring particles replace the parent particles; the position and speed of the offspring particles are calculated as follows:

[0070]

[0071] In the formula, p is a random number between 0 and 1, child(x) represents the position of the child particle, child(v) represents the speed of the child particle, parent1(x), parent2(x), parent1(v), parent2(v) represent the position and speed of the parent particle respectively. After the crossover operation, the same child individuals as the parent are generated, ensuring the consistency of the number of individuals in the particle swarm algorithm. At the same time, the diversity of the population is increased, individuals with poor fitness are eliminated, and the convergence speed of particles is accelerated.

[0072] The adaptive particle mutation operation introduced in the present invention enhances the search capability of particles in the later stage. That is, after each update, the particles are reinitialized according to the mutation frequency pm, and pm is set between 0 and 1. When the particle swarm algorithm evolves to the later stage, its search space gradually becomes smaller. The mutation operation helps particles expand the search space in the later stage of evolution, and can help particles jump out of the optimal position it has searched before, and then search in a larger range, which can not only maintain the diversity of the population, but also improve the possibility of the algorithm finding the optimal value.

[0073] The present invention uses the SVM algorithm to train the model in detail, and the method is as follows:

[0074] The present invention uses the libsvmtrain function of the LIBSVM toolbox to train the sample, model = libsvmtrain (trainlabel, trainfinal, ['options']). Wherein model is the obtained training result, which is a structure, that is, a training model; trainlabel is a known category label, which is an N*1 matrix; trainfinal is the sample data we input, that is, the training set sample mentioned above (the area, height, width, centroid and number of connected regions of the binary image obtained by flame segmentation and feature extraction and the combination of other fire parameters), which is an N*M numerical matrix, N is the number of sample points, and M is the characteristic dimension of the sample; ['options'] is the optional operation parameter of the libsvmtrain function, including the category of the kernel function, the penalty parameter c and the kernel parameter g value. The classification kernel function selected by the present invention is the RBF kernel function, and the parameters c and g are determined by the optimization function.

[0075] Combined with the above-mentioned improvement ideas, the GA-PSO algorithm is used for optimization, and it is named as the improved particle swarm algorithm-support vector machine model. The specific steps of the fire combustion form recognition model proposed by the present invention are as follows:

[0076] (1) Randomly initialize the speed, position, and all related parameters of the population particles in the input training set sample data. Calculate the initial fitness values ​​of all particles so that the particles can find the optimal solution for the flame combustion form identification parameters;

[0077] (2) Find the initial extreme value, determine the individual extreme value pbest and the group extreme value gbest according to the fitness value of the initial particles, prevent only finding the local optimal solution of the flame combustion form identification parameters, and facilitate the particles to gather roughly at the optimal solution;

[0078] (3) Update the speed and position of the offspring particles according to the proposed crossover formula and replace the parent particle position to further capture flame combustion data, such as the center of mass position and area;

[0079] (5) Perform adaptive particle mutation;

[0080] (6) Update inertia weight;

[0081] (7) If the stopping condition is met or the maximum number of iterations is exceeded, the optimal solution is output; otherwise, jump to step (3) to continue searching.

[0082] In summary, Python programming is used to binarize the images obtained by video frame segmentation, perform flame area segmentation and feature extraction, merge the obtained data with other fire parameters, mark the combustion form to construct a data set, and divide it into training set and test set in a ratio of 7:3. The samples of the training set are input into the pre-established improved particle swarm algorithm-support vector machine model for training, and then the classification labels are predicted through the test set, and finally the classification and recognition results of the combustion form of the substation oil filling equipment fire are obtained.

[0083] Example 2

[0084] Embodiment 2 of the present invention provides a system for identifying the combustion form of oil-filled equipment fire in a substation, the system being used by the aforementioned method for identifying the combustion form of oil-filled equipment fire in a substation, comprising:

[0085] Image acquisition module, used to collect surveillance camera videos during the substation fire and experimental images of separate combustion forms;

[0086] Flame area segmentation module, used to obtain the flame combustion area;

[0087] A flame feature extraction module is used to extract image features of the flame combustion area;

[0088] The combustion form recognition module is used to classify and recognize the flame combustion form.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

[0090] The beneficial effects of the present invention are that, compared with the prior art, the present invention has the following advantages:

[0091] (1) Fully combine the data of the monitoring equipment in the station, such as the pressure inside the equipment, the oil temperature of the oil-filled equipment, the side wall temperature, the flame temperature, the smoke composition and the image feature parameters (area, height, width, centroid and connected area, etc.). After expanding the data dimension, the algorithm can more accurately identify the combustion form in the early stage of the fire, and the false alarm rate is lower.

[0092] (2) The structure of the particle swarm algorithm (PSO) is relatively simple, easy to operate and implement, with a fast convergence speed, and no complex operations such as selection, crossover and mutation. However, the particle swarm algorithm has the problems of easy premature convergence, relatively low classification accuracy, and easy to fall into the local optimal solution. In order to avoid the above problems in the identification of the combustion form of substation fire, the present invention combines the advantages of the PSO and GA algorithms, and proposes a hybrid optimization algorithm (GA-PSO) based on the adaptive weighted particle swarm algorithm and introduces the crossover and mutation operator of the genetic algorithm (GA) to improve the search performance of the particle swarm algorithm. The GA-PSO algorithm maximizes the advantages of the two algorithms and avoids the shortcomings of the particle swarm algorithm. The algorithm increases the diversity of the population and increases the original search range of the particle swarm algorithm, improving the defects of the particle swarm algorithm in the search process, such as premature convergence and easy to fall into the local optimal value.

[0093] (3) The method for identifying the combustion form of oil-filled equipment fire in a substation of the present invention can timely identify the combustion form of the substation fire in its early stage through the improved particle swarm-support vector machine model, and generate the fire burning area, height, width and combustion form in real time. The algorithm has high optimization efficiency and strong recognition performance, and can effectively identify and classify the combustion form of flames, and assist firefighters in fire departments to carry out targeted firefighting operations.

[0094] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0095] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0096] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0097] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for identifying the combustion form of oil-filled equipment fire in a substation, characterized in that: include: Image acquisition: Collect surveillance camera videos and experimental images of individual combustion during the substation fire, and use video framing technology to extract the flame image per second in the video; Flame area segmentation: The images obtained by dividing the above video frames are binarized through Python programming. Based on the image binarization technology, an adaptive threshold function is introduced to process the original image of the data set to obtain the corresponding flame burning area; Flame feature extraction: Based on the connected region detection method, the area, height, width, centroid and number of connected regions of the binary image of the flame combustion region are calculated; Acquisition of other fire parameters: Collect other parameters that characterize the combustion form of oil-filled equipment fires, mainly including temperature, smoke composition and internal pressure of the equipment; Dataset construction steps: The area, height, width, centroid and number of connected regions of the binary image of the flame burning area at the same time are combined with other fire parameters, and the combustion form is marked. The oil pool fire is marked as 1, the flowing fire is marked as 2, and the jet fire is marked as 3. Finally, it is divided into training set and test set in a ratio of 7:3; Combustion form identification: the data of the training set samples are input into a pre-established and trained improved particle swarm algorithm-support vector machine model to predict the classification labels and obtain the classification identification results of the flame combustion form.

2. A method for identifying the combustion form of oil-filled equipment fire in a substation according to claim 1, characterized in that: The connected regions are regions with the same pixel values ​​and adjacent to each other. The number of connected regions is used to describe the splitting of the flame. A plurality of connected regions indicates that the flame has a plurality of burning points or the shape of the flame is irregular and the flame burns in a dispersed manner.

3. A method for identifying the combustion form of oil-filled equipment fire in a substation according to claim 2, characterized in that: The other flame parameter extractions include: The temperature includes oil temperature, external axial temperature of the oil-filled equipment and side wall temperature, which can be obtained by connecting a data acquisition instrument with a K-type thermocouple, and directly obtaining the temperature and heat flux density data signals as one-dimensional data; the internal pressure of the oil-filled equipment can be obtained through a pressure transmitter; the changes in the flue gas composition can be obtained through a flue gas collector to monitor the changes in the CO and CO2 content.

4. A method for identifying the combustion form of oil-filled equipment fire in a substation according to claim 3, characterized in that: The combustion form identification is based on the adaptive weighted particle swarm algorithm and introduces the GA-PSO algorithm. In each iteration, the fitness values ​​of the particles are sorted from large to small. The larger the fitness value of the particle, the more adaptable it is. The fitness function is the average value of the classification accuracy under 3-fold cross validation. First, the particles with higher fitness values ​​are selected to directly enter the next generation to ensure that the next generation of particles is relatively better; at the same time, the positions and speeds of the first half of the particles with higher fitness values ​​are assigned to the second half of the particles with lower fitness values. These particles are randomly crossed in pairs, and then the same number of offspring particles are generated, and the offspring particles replace the parent particles.

5. A method for identifying the combustion form of oil-filled equipment fire in a substation according to claim 4, characterized in that: The position and velocity of the offspring particles are calculated as follows: Where p is a random number between 0 and 1, child(x) represents the position of the child particle, child(v) represents the velocity of the child particle, parent1(x), parent2(x), parent1(v), parent2(v) represent the position and velocity of the parent particle respectively.

6. A method for identifying the combustion form of oil-filled equipment fire in a substation according to claim 5, characterized in that: The SVM algorithm is introduced into the combustion form recognition process to carry out specific training of the model, including: The samples are trained using the libsvmtrain function of the LIBSVM toolbox, model = libsvmtrain(trainlabel, trainfinal, ['options']), where model is the obtained training model; trainlabel is the known category label, which is an N*1 matrix; trainfinal is the input sample data, which is the area, height, width, centroid and number of connected regions of the binary image obtained by flame segmentation and feature extraction, combined with other fire parameters, and is an N*M numerical matrix, where N is the number of sample points and M is the feature dimension of the sample; ['options'] is the optional operation parameter of the libsvmtrain function, including the category of the kernel function, the penalty parameter c and the kernel parameter g value. The selected classification kernel function is the RBF kernel function, and the parameters c and g are determined by the optimization function.

7. A method for identifying the combustion form of oil-filled equipment fire in a substation according to claim 6, characterized in that: The working process of the flame combustion form identification model includes: (1) In the input training set sample data, the speed, position, and all related parameters of the population particles are randomly initialized; the initial fitness values ​​of all particles are calculated so that the particles can find the optimal solution for the flame combustion form identification parameters; (2) Find the initial extreme value, determine the individual extreme value pbest and the group extreme value gbest according to the fitness value of the initial particles, prevent only finding the local optimal solution of the flame combustion form identification parameters, and facilitate the particles to gather roughly at the optimal solution; (3) Update the speed and position of the offspring particles according to the proposed crossover formula and replace the parent particle position to further capture the flame combustion data; (5) Perform adaptive particle mutation; (6) Update inertia weight; (7) If the stopping condition is met or the maximum number of iterations is exceeded, the optimal solution is output; otherwise, jump to step (3) to continue searching.

8. A system for identifying the combustion form of oil-filled equipment fire in a substation, the system being the system used in the method for identifying the combustion form of oil-filled equipment fire in a substation as claimed in any one of claims 1 to 7, characterized in that: include: Image acquisition module, used to collect surveillance camera videos during the substation fire and experimental images of separate combustion forms; Flame area segmentation module, used to obtain the flame combustion area; A flame feature extraction module is used to extract image features of the flame combustion area; The combustion form recognition module is used to classify and recognize the flame combustion form.

9. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images