A system and method for tracing potential hazards of current overload of a metallic material

Abstract

The application discloses a hidden danger tracing system and method for metal material current overload, utilizes the characteristics that micro-particles and gas are generated by overheating decomposition of marking materials coated on the surface of metal high-temperature components, matches the heating decomposition curve of the marking materials by identifying and analyzing the conditions of the micro-particles and the gas, obtains the curve estimation result of the current safety condition of the metal high-temperature equipment, simultaneously inputs the sampling gas data into the trained BP neural network model, estimates the current safety condition of the metal high-temperature equipment by using the intelligent analysis algorithm of the BP neural network as the BP network estimation result, combines the curve estimation with the BP neural network estimation to optimize the current safety condition of the metal high-temperature equipment, determines the early warning level of the fire hidden danger, realizes the graded early warning of the fire safety accidents, and is earlier than the early warning time of the traditional smoke sensing or temperature sensing fire early warning mode, and can obtain more rescue time for operation and maintenance rescue personnel.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fire safety tracking and early warning in the scenario of heat accumulation caused by heat production of high-temperature metal equipment, and particularly relates to a hidden danger tracking system and method for current overload of metal material. BACKGROUND

[0002] High-temperature metal equipment is stable in material quality, and the temperature continuously rises during operation. When the high-temperature metal equipment is overheated due to overcurrent or overload and the heat dissipation capacity is insufficient, heat accumulation is easily caused, the temperature in the space continuously accumulates, and a fire occurs. Especially in a confined space, the ventilation is poor, and when the space has insufficient heat dissipation capacity, there is a serious fire hazard.

[0003] At the same time, the confined space with high-temperature equipment inside is usually located in a remote area, and the operation and rescue personnel often cannot arrive in time to deal with it, which easily causes serious economic losses. For example, in a wind turbine, in addition to electrical equipment, there are also a large number of heat-producing components such as gearboxes and generators in the wind turbine cabin. The heat dissipation capacity in the wind turbine cabin is insufficient, which easily causes the heat dissipation of these heat-producing components to be insufficient, and then the local temperature in the cabin is too high, and a fire is caused.

[0004] For the situation of fire hazard caused by heat accumulation due to heat production of high-temperature metal equipment, the existing technology still mainly uses traditional fire monitoring. Especially in a confined space with poor ventilation, the current commonly used fire hazard monitoring methods include smoke monitoring (judging whether there is a fire hazard in the environment according to the smoke in the environment), flame monitoring (evaluating the fire hazard according to whether there is a flame in the environment), gas monitoring (evaluating the fire hazard according to the CO in the environment), video monitoring and the like.

[0005] However, in actual production and life, the mainstream monitoring methods such as smoke detection, flame monitoring and gas monitoring all have certain disadvantages in early warning time. For example, smoke monitoring can only monitor the stage when visible smoke exists in the environment, flame monitoring can only monitor the stage when a flame appears, and gas monitoring mainly monitors the gas diffused from the hidden danger source to the monitoring point. Smoke monitoring and flame monitoring cannot monitor the early stage of a fire, gas monitoring has the problem of insufficient timeliness, and video monitoring is difficult to achieve timely early warning before a fire occurs.

[0006] Moreover, in a confined space, operation and rescue personnel cannot go in and out at will, and it takes a lot of time to arrive at and enter the scene each time. The existing mainstream monitoring methods cannot completely solve the urgency of fire rescue in time. For example, in a wind farm scene, wind turbines are usually located in a relatively remote area, and the smoke, flame and gas monitoring have the problem of not timely early warning, which greatly compresses the rescue time of operation personnel.

[0007] Therefore, in view of the deficiency in the fire hazard monitoring technology for high-temperature metal components in a confined space, a system and method capable of early warning before heat generated by current overload of a metal high-temperature device is accumulated and a fire occurs are needed. SUMMARY

[0008] To solve the above technical problems, the present application provides a metal material current overload hazard tracking system, which can prevent further spread of fire, issue an early warning signal in time before the fire occurs, and gain more rescue time for operation and maintenance rescue personnel.

[0009] Meanwhile, the present application also provides a metal material current overload hazard tracking method, which can issue an early warning in time before the fire occurs and gain more rescue time for operation and maintenance rescue or fire department.

[0010] The technical solution of the present application is as follows: a metal material current overload hazard tracking system, comprising:

[0011] A material marking module is configured to apply numbered marking materials on the surface of a specified position of a metal high-temperature device.

[0012] An information preset module is configured to draw a thermal decomposition curve of the marking materials by isothermal gradient heating and match the curve with the number of the marking materials.

[0013] A space sampling module is configured to sample gas in the surrounding space of the specified position of the metal high-temperature device and input the sampling data into a curve estimation module and a feature analysis module.

[0014] The curve estimation module is configured to extract and analyze the data of the sampled micro-particles and characteristic gas, match the data with the thermal decomposition curve of the marking materials, and obtain a curve estimation result of the current safety condition of the metal high-temperature device.

[0015] The feature analysis module is configured to input the data of the sampled gas into a trained BP neural network model and use a BP neural network intelligent analysis algorithm to estimate the current safety condition of the metal high-temperature device as a BP network estimation result.

[0016] An estimation optimization module is configured to weight and average the BP network estimation result and the curve estimation result according to a set weight ratio to obtain a final estimation result of the current safety condition of the metal high-temperature device.

[0017] A warning output module is configured to determine a warning level of a fire hazard according to the final estimation result of the current safety condition of the metal high-temperature device and output the warning level.

[0018] The metal material current overload hidden danger tracing system, wherein the BP neural network model adopted by the feature analysis module is a three-layer BP neural network model, including:

[0019] A sampling parameter input layer is arranged with 2n neuron nodes x 11 , x 12 , x 21 , x 22 , …, x n1 and x n2 , which respectively correspond to the data of 2n sampled gases, including gas 1 concentration, gas 1 concentration change rate, gas 2 concentration, gas 2 concentration change rate, …, gas n-1 concentration, gas n-1 concentration change rate, micro-particle concentration and micro-particle concentration change rate; wherein n represents a natural number;

[0020] A function iteration layer is arranged with 3 neuron nodes, which respectively correspond to a first function f(u) = 1 / (1+e u ), a second function f(u) = (e u -e -u ) / (e u +e -u ) and a third function f(u) = max(0, u);

[0021] A warning level output layer is arranged with 1 neuron node corresponding to a fourth function f(x) = (e x -e -x ) / (e x +e -x ).

[0022] The metal material current overload hidden danger tracing system, wherein the gas 1 concentration change rate, gas 2 concentration change rate, …, gas n-1 concentration change rate and micro-particle concentration change rate are the concentration change values of the gas or micro-particle per minute; the particle size of the micro-particle is between 2 nm and 20 μm.

[0023] The metal material current overload hidden danger tracing system, wherein the marking material is polyvinyl chloride, polyamide fiber, chlorobutyl rubber, polyurethane or polyvinyl chloride amide; the sampled gas is H2, CO, HCL, CO2, VOCs and various hydrocarbon gases.

[0024] The metal material current overload hidden danger tracing system, wherein the application place of the system is a wind turbine cabin, a switch cabinet or an energy storage cabin.

[0025] A metal material current overload hidden danger tracing method, including the following steps:

[0026] A, coating a marking material on the surface of a designated position of a metal high-temperature equipment and numbering;

[0027] B, obtain the number of the marking material, and call the thermal decomposition curve corresponding to the number of the marking material;

[0028] C, gas sampling is performed on the surrounding space at the specified position of the metal high-temperature equipment;

[0029] D, data of the sampled micro-particles and characteristic gas are extracted and analyzed, and are matched with the thermal decomposition curve of the marking material, to obtain a curve estimation result of the current safety condition of the metal high-temperature equipment;

[0030] E, data of the sampled gas are input into the trained BP neural network model, and a BP neural network intelligent analysis algorithm is used to estimate the current safety condition of the metal high-temperature equipment as a BP network estimation result;

[0031] F, the BP network estimation result and the curve estimation result are weighted and averaged according to a set weight ratio, to obtain a final estimation result of the current safety condition of the metal high-temperature equipment;

[0032] G, a fire hazard warning level is determined according to the final estimation result of the current safety condition of the metal high-temperature equipment, and is output.

[0033] The metal material current overload hazard tracing method, wherein the BP neural network intelligent analysis algorithm in step E includes the following specific steps:

[0034] E21, sampling parameter input layer parameters x = [x 11 , x 12 , x 21 , x 22 , …, x n1 and x n2 ] are set;

[0035] E22, the sampling parameter input layer parameters x are input into a function iteration layer composed of a first function f(u) = 1 / (1+e u ), a second function f(u) = (e u -e -u ) / (e u +e -u ) and a third function f(u) = max(0, u), and are calculated together with known parameters generated by automatic iteration in the trained model;

[0036] E23, the calculation result of step E2 is input into a warning level output layer composed of a fourth function f(x) = (e x -e -x ) / (e x +e -x ), to obtain the value of output y = f(x);

[0037] E24, determining the fire hazard warning level according to the calculation result of step E3.

[0038] The metal material current overload hazard tracking method, wherein, in step E24, the numerical value according to y=f(x) is used to determine the fire hazard warning level: when y∈[0, 0.15), it represents safety; when y∈[0.15, 0.3), it represents a first-level warning; when y∈[0.3, 0.45), it represents a second-level warning; when y∈[0.45, 0.6), it represents a third-level warning; when y∈[0.6, 0.75), it represents a fourth-level warning; and when y∈[0.75, 1), it represents a fifth-level warning.

[0039] The metal material current overload hazard tracking method, wherein, in step E, the specific steps for training the BP neural network model are as follows:

[0040] E11, preparing and arranging the input vector X and the target vector P for training the BP neural network model; the input vector X is a data set composed of the concentrations of gas 1, gas 1 concentration rate, gas 2, gas 2 concentration rate, …, gas n-1, gas n-1 concentration rate, micro-particle concentration, and micro-particle concentration rate; and the target vector P is the fire hazard warning level corresponding to each parameter range;

[0041] E12, substituting the input vector X into the BP neural network model and iteratively calculating with the known parameters;

[0042] E13, calculating the error between the output y and the target vector P;

[0043] E14, determining whether the error falls between 0.01 and 0.1; if yes, stopping the training; otherwise, adjusting the numerical value of the known parameters and returning to step E12 for repeated iteration and update.

[0044] The metal material current overload hazard tracking method, wherein, in step F, the BP network prediction result accounts for 60% of the weight, and the curve prediction result accounts for 40% of the weight.

[0045] The metal material current overload hazard tracking system and method provided by the application utilize the characteristics of micro-particle products and gas products generated by the overheating decomposition of the marking material coated on the surface of the metal high-temperature component, identify and analyze the conditions of the micro-particle products and gas products, combine curve prediction with BP neural network prediction to optimize the current safety status of the metal high-temperature equipment, determine the warning level of the fire hazard, realize early-stage graded warning of fire safety accidents, and have earlier warning time than the traditional smoke-sensing or temperature-sensing fire warning method, which can help the operation and maintenance rescue personnel to gain more rescue time. BRIEF DESCRIPTION OF DRAWINGS

[0046] The drawings described herein are for purposes of illustration only and are not intended to limit the scope of the present disclosure in any way; the shapes and relative sizes and dimensions of the various components in these figures are meant only to be illustrative - specific shapes and relative sizes and dimensions are not intended to limit the scope of the present disclosure, and various changes in the shapes and relative sizes and dimensions of the components can be made without departing from the spirit of the present disclosure.

[0047] Figure 1 is the constituent block diagram of the metal material current overload fire hazard tracing system embodiment of the present disclosure;

[0048] Figure 2 is the thermal decomposition curve diagram of the micro-particle of polyvinyl chloride used in the metal material current overload fire hazard tracing system embodiment of the present disclosure;

[0049] Figure 3 is the thermal decomposition curve diagram of the characteristic gas of polyvinyl chloride used in the metal material current overload fire hazard tracing system embodiment of the present disclosure;

[0050] Figure 4 is the structural schematic diagram of the space sampling module used in the metal material current overload fire hazard tracing system embodiment of the present disclosure;

[0051] Figure 5 is the principle block diagram of the multi-layer BP neural network model used in the metal material current overload fire hazard tracing system embodiment of the present disclosure;

[0052] Figure 6 is the neuron node schematic diagram of the multi-layer BP neural network model used in the metal material current overload fire hazard tracing system embodiment of the present disclosure;

[0053] Figure 7 is the flow block diagram of the metal material current overload fire hazard tracing method embodiment of the present disclosure.

[0054] The labels in the drawings are summarized: material marking module 110, information preset module 120, space sampling module 130, sampling tube 131, sampling fan 132, curve estimation module 140, characteristic analysis module 150, sampling parameter input layer 151, function iteration layer 152, early warning level output layer 153, estimation optimization module 160, early warning output module 170. DETAILED DESCRIPTION

[0055] The specific embodiments and examples of the present disclosure will be described in detail below with reference to the accompanying drawings, and the specific embodiments described are only used to explain the present disclosure, and are not intended to limit the specific embodiments of the present disclosure.

[0056] like Figure 1 As shown, the hazard tracing system for current overload of metallic materials of the present invention includes a material marking module 110, an information preset module 120, a spatial sampling module 130, a curve prediction module 140, a feature analysis module 150, a prediction optimization module 160, and an early warning output module 170.

[0057] The material marking module 110 is used to coat the surface of a metal high-temperature device with a numbered marking material at a designated location. This is especially useful for parts of the metal high-temperature device in a confined space that are prone to high temperatures due to current overload of the metal material, such as gearboxes and engines of wind turbines. As the operating temperature gradually increases, the marking material will release gaseous and particulate products to varying degrees. The release of these gaseous and particulate products will also vary depending on the structure, temperature, and humidity of the monitoring environment. By utilizing the overheating decomposition property of the marking material, the high-temperature metal components in the confined space can be monitored. When the temperature of the heat-generating component is too high, the marking material coated on the surface of the heat-generating component will decompose to produce particulate and gaseous products.

[0058] The labeling materials mentioned in this article refer to: polyvinyl chloride (PVC), polyamide fiber (nylon), chloroprene rubber (CR), polyurethane (TPU), or polyvinyl chloride amide; gaseous products refer to characteristic gases such as H2, CO, HCl, CO2, VOCs, and various hydrocarbon gases; microparticle products refer to microparticles with a particle size between 2nm and 20μm. In other words, the thermal decomposition concentration and particle size will be different under temperature changes when different labeling materials are applied, which can help diagnose the temperature status of metal equipment on site. Taking polyvinyl chloride as an example, the characteristic gases of its thermal decomposition are VOC, H2, and CO.

[0059] The tracing mentioned in this article is also called monitoring; the high-temperature metal equipment mentioned in this article can be a gearbox and / or generator in a wind turbine nacelle, a switch cabinet, or a battery cell and / or battery pack casing in an energy storage compartment; the designated location refers to the location where the high-temperature metal equipment has a relatively concentrated heat or is prone to heat accumulation during operation; the coating method is mainly local spot coating or local scribing, which will not affect the heat dissipation performance of the gearbox and generator.

[0060] The information preset module 120 is used to plot the thermal decomposition curve of the marked material using isothermal gradient heating, and to correspond one-to-one with the number of the marked material; taking polyvinyl chloride as an example, after isothermal gradient heating, the changes in the concentration of microparticles and the concentration changes of characteristic gases are as follows: Figure 2 and Figure 3 As shown, Figure 2 It is a curve showing the change in microparticle concentration over time. Figure 2It can be clearly seen from the figure that the concentration of the micro-particles with a particle size of 0.3 μm changes greatly over time, Figure 3 is a graph of the concentration of the characteristic gas over time, from Figure 3 It can also be clearly seen from the figure that the concentration of the characteristic gas VOC changes more dramatically over time.

[0061] The space sampling module 130 is configured to sample the gas in the surrounding space of the designated position of the metal high-temperature equipment and input the sampling data into the curve estimation module 140 and the characteristic analysis module 150.

[0062] Taking the gear box and the generator in the nacelle of the wind turbine as an example, as shown in Figure 4 , specifically, the space sampling module 130 includes a sampling tube 131 and a sampling fan 132. The sampling tube 131 is arranged in the confined space of the nacelle of the wind turbine and located above the metal high-temperature equipment of the gear box and the generator. The sampling fan 132 is arranged at each air inlet of the sampling tube 131. The air inlets of the sampling tube 131 are located above the designated position coated with the marking material.

[0063] The curve estimation module 140 is configured to extract and analyze the data of the sampled micro-particles and gas, match the data with the thermal decomposition curve of the marking material, and obtain the curve estimation result of the current safety condition of the metal high-temperature equipment.

[0064] Taking the gear box and the generator in the nacelle of the wind turbine coated with polyvinyl chloride as an example, the curve estimation module 140 extracts and analyzes the data of the sampled micro-particles and characteristic gas, specifically including the particle size and concentration of the micro-particles, the identification of the gas type, and the concentration of the gas, and matches the data with the thermal decomposition curve of polyvinyl chloride, from Figure 2 which the current safety condition, i.e., the safety level, of the gear box and the generator in the nacelle of the wind turbine can be obtained.

[0065] The characteristic analysis module 150 is configured to input the data of the sampled gas into the trained BP neural network model and use the BP neural network intelligent analysis algorithm to estimate the current safety condition of the metal high-temperature equipment as the BP network estimation result.

[0066] As shown in Figure 5 , the input of the multi-layer BP neural network model includes the concentration of gas 1, the concentration change rate of gas 1, the concentration of gas 2, the concentration change rate of gas 2, …, the concentration of gas n-1, the concentration change rate of gas n-1, the concentration of micro-particles, and the concentration change rate of micro-particles. The output includes safety, first-level warning, second-level warning, third-level warning, fourth-level warning, and fifth-level warning.

[0067] Taking the three-layer BP neural network model as an example, as shown in Figure 6 , it includes a sampling parameter input layer 151, a function iteration layer 152, and a warning level output layer 153.

[0068] The sampling parameter input layer 151 is set with 2n neuron nodes x 11 x 12 x 21 x 22 ... x n1 and x n2 , respectively corresponding Figure 5 The data includes the concentration of gas 1, the rate of change of gas 1 concentration, the concentration of gas 2, the rate of change of gas 2 concentration, ..., the concentration of gas n-1, the rate of change of gas n-1 concentration, the concentration of microparticles, and the rate of change of microparticle concentration for 2n sampled gases; where n represents a natural number.

[0069] Specifically, the concentration change rate of gas 1, the concentration change rate of gas 2, ..., the concentration change rate of gas n-1 and the concentration change rate of microparticles are the concentration changes of gas or microparticles per minute; the particle size of the microparticles is between 2 nm and 20 μm.

[0070] Function iteration layer 152 has 3 neuron nodes b 13 b 23 and b 33 , respectively corresponding to:

[0071] The first function is f(u) = 1 / (1+e) u );

[0072] The second function f(u) = (e u -e -u ) / (e u +e -u );

[0073] The third function is f(u) = max(0,u);

[0074] In the first function, u is the calculation result u1 of the normalized one-dimensional index vector (hereinafter referred to as the index vector) established with various gas concentrations in the preset model. u1 serves as the neuron node b. 13 The input variables are substituted into the first function to calculate f1; u in the second function is the calculation result u2 of the input feature vector in the preset model, and u2 is used as the neuron node b. 23 The input variables are substituted into the second function to calculate f2; the u in the third function is the calculation result u3 of the input feature vector in the preset model, and u3 is used as the neuron node b. 33 Substitute the input variables into the third function to calculate f3;

[0075] Here, we select 3 neuron nodes b. 13 b 23 and b 33And different functions are adopted, the disadvantages of different functions can be balanced, the gradient loss in the calculation process of the BP neural network model is prevented, and the prediction accuracy and calculation efficiency can be obviously improved.

[0076] The early warning level output layer 153 is provided with one neuron node b 43 Corresponding to:

[0077] The fourth function f(x) = (e x -e -x ) / (e x +e -x );

[0078] Wherein, x in the fourth function is the calculation result u4 of the preset model, taking the calculation results of the first, second and third functions as a feature vector [f1, f2, f3], u4 as an input variable of the neuron node b 43 , and y is calculated, and the value of y can be used for safety level estimation.

[0079] The estimation optimization module 160 is used for weighting and averaging the BP network estimation result and the curve estimation result according to the set weight proportion, so as to obtain the final estimation result of the current safety status of the metal high-temperature equipment, so as to prevent the failure of one of the estimation results from causing the distortion of the final estimation result.

[0080] Preferably, the BP network estimation result accounts for 60% of the weight, and the curve estimation result accounts for 40% of the weight; for example, the curve estimation result is 1 (representing a first early warning), and the BP network estimation result is 2 (representing a second early warning), then the final estimation result is 1*40%+2*60%=0.4+1.2=1.6≈2, that is, the final estimation result is a second early warning.

[0081] The early warning output module 170 is used for determining the early warning level of the fire hazard according to the final estimation result of the current safety status of the metal high-temperature equipment and outputting; the early warning level can be divided into safety, first early warning, second early warning, third early warning, fourth early warning and fifth early warning; the first early warning is the earliest early warning level of the fire hazard.

[0082] Based on the above-mentioned metal material current overload hazard tracking system, the application further provides a metal material current overload hazard tracking method, as shown in Figure 7 The method comprises the following steps:

[0083] Step S210, coating a marking material on a specified position of the surface of the metal high-temperature equipment and numbering;

[0084] Step S220, acquiring the number of the marking material, and calling the heat decomposition curve corresponding to the number of the marking material;

[0085] Step S230, gas sampling is performed on the peripheral space at the designated position of the metal high-temperature equipment;

[0086] Step S240, data of the sampled micro-particles and characteristic gas are extracted and analyzed, and are matched with the thermal decomposition curve of the marking material to obtain a curve estimation result of the current safety condition of the metal high-temperature equipment;

[0087] Step S250, the data of the sampled gas are input into the trained BP neural network model, and a BP neural network intelligent analysis algorithm is used to estimate the current safety condition of the metal high-temperature equipment as a BP network estimation result;

[0088] Step S260, the BP network estimation result and the curve estimation result are weighted and averaged according to a set weight ratio to obtain a final estimation result of the current safety condition of the metal high-temperature equipment; on one hand, the estimation is matched according to the thermal decomposition curve of the marking material, and on the other hand, the estimation is classified according to the BP neural network intelligent algorithm, and the final conclusion is obtained by combining the estimation results of the two methods.

[0089] Step S270, the final estimation result of the current safety condition of the metal high-temperature equipment is used to determine the warning level of the fire hazard and output.

[0090] Specifically, the BP neural network intelligent analysis algorithm in step S250 includes the following steps:

[0091] S255, setting the sampling parameter input layer 151 parameters x = [x1, x2, x3, x4, x5, x6]; 11 12 21 22 n1 n2

[0092] S256, inputting the sampling parameter input layer 151 parameters x into a function iteration layer 152 composed of a first function f(u) = 1 / (1+e u ), a second function f(u) = (e u -e -u ) / (e u +e -u ) and a third function f(u) = max(0, u), and performing calculation together with known parameters generated by automatic iteration in the trained model;

[0093] Taking the sampling parameter input layer 151 parameters x = [x1, x2, x3, x4, x5, x6] as an example, the input quantities u1, u2 and u3 of the function iteration layer 152 are calculated respectively by bringing them into the preset model, and the calculation method is as follows:

[0094] u1 = w 11 x1+w​​​​​​12 x2+w 13 x3+w 14 x4+w 15 x5+w 16 x6+b1;

[0095] u2=w 21 x1+w 22 x2+w 23 x3+w 24 x4+w 25 x5+w 26 x6+b2;

[0096] u3=w 31 x1+w 32 x2+w 33 x3+w 34 x4+w 35 x5+w 36 x6+b3;

[0097] wherein, w 11 , w 12 , w 13 , w 14 , w 15 , w 16 , b1; w 21 , w 22 , w 23 , w 24 , w 25 , w 26 , b2; w 31 , w 32 , w 33 , w 34 , w 35 , w 36 , b3 are known parameters generated after automatic iteration in the trained model;

[0098] The output quantities f1, f2, f3 of the function iteration layer 152 are calculated again, and the calculation method is as follows:

[0099] f1=1 / (1+e u1 );

[0100] f2=(e u2 -e -u2 ) / (e u2 +e -u2 )

[0101] f3=max(0, u3);

[0102] S257, the calculation result of step S256 is input into the fourth function f(x)=(e x -e-x ) / (e x +e -x The output layer 153, composed of warning levels, is used to calculate the value of the output quantity y = f(x).

[0103] Taking f1, f2, and f3 in step S256 as an example, first calculate the input u4 = w of the warning level output layer 153. 41 f1+w 42 f2+w 43 f3+b4, where w 41 w 42 w 43 b4 are also known parameters generated automatically after iteration in the trained model;

[0104] Next, calculate the output quantity y = f(u4) = (e) of the warning level output layer 153. u4 -e -u4 ) / (e u4 +e -u4 );

[0105] S258. Determine the fire hazard warning level based on the calculation results of step S257.

[0106] Specifically, in step S258, the fire hazard warning level is determined according to the value of y = f(x) within the following range: when y ∈ [0, 0.15), it represents safety; when y ∈ [0.15, 0.3), it represents a level one warning; when y ∈ [0.3, 0.45), it represents a level two warning; when y ∈ [0.45, 0.6), it represents a level three warning; when y ∈ [0.6, 0.75), it represents a level four warning; and when y ∈ [0.75, 1), it represents a level five warning.

[0107] Here, the use of y values ​​to determine the warning level is different from the ordinary direct classification method, i.e., the direct classification in the BP neural network. If there are too many categories, it is easy to cause excessive computation. The predicted values ​​used here can be further processed and classified according to the magnitude of the predicted values, which makes the calculation faster. Moreover, the values ​​of y are all between 0 and 1, and all data are within the same measurement scale, which is also easier to process.

[0108] Specifically, the steps for training the BP neural network model in step S250 are as follows:

[0109] S251, prepare and arrange the input vector X and the target vector P for training the BP neural network model; the input vector X is a data set composed of the concentration of gas 1, the concentration change rate of gas 1, the concentration of gas 2, the concentration change rate of gas 2, …, the concentration of gas n-1, the concentration change rate of gas n-1, the concentration of micro-particles, and the concentration change rate of micro-particles; the target vector P is the fire hazard early warning level corresponding to each parameter range;

[0110] S252, substitute the input vector X into the BP neural network model, and perform iterative calculation together with the known parameters trained;

[0111] S253, calculate the error between the output y and the target vector P;

[0112] S254, determine whether the error falls within 0.01-0.1, if yes, stop training; otherwise, adjust the values of the known parameters (i.e., similar to the aforementioned w 11 , w 12 , w 13 , w 14 , w 15 , w 16 , b1; w 21 , w 22 , w 23 , w24, w 25 , w 26 , b2; w 31 , w 32 , w 33 , w 34 , w 35 , w 36 , b3; w 41 , w 42 , w 43 , b4) and return to step E12 for repeated iteration and update until the error value between the output y and the target vector P is within the allowable range of 0.01-0.1; finally obtain the known parameters generated by automatic iteration in the trained model required in step S256.

[0113] The contents not described in detail in the specification all belong to the prior art known to those skilled in the art.

[0114] It should be understood that the above description is only the preferred embodiment of the present application and is not intended to limit the technical solutions of the present application. Those skilled in the art can make additions, substitutions, transformations or improvements to the above description within the spirit and principles of the present application, and all these additions, substitutions, transformations or improvements shall fall within the protection scope of the appended claims of the present application.

Claims

1. A metal material current overload hazard tracking system, characterized by, The system comprises: a material marking module for coating a numbered marking material on the surface of a specified position of a metal high-temperature equipment; an information preset module for drawing a thermal decomposition curve of the marking material by an isothermal gradient heating method, and corresponding to the number of the marking material; a space sampling module for sampling gas in the surrounding space of the specified position of the metal high-temperature equipment, and inputting the sampling data into a curve estimation module and a feature analysis module; the curve estimation module for extracting and analyzing the data of the sampled micro-particles and characteristic gas, matching the thermal decomposition curve of the marking material, and obtaining a curve estimation result of the current safety condition of the metal high-temperature equipment; the feature analysis module for inputting the data of the sampled gas into a trained BP neural network model, and using a BP neural network intelligent analysis algorithm to estimate the current safety condition of the metal high-temperature equipment as a BP network estimation result; the BP neural network model is a three-layer BP neural network model, comprising: A sampling parameter input layer is provided with 2n neuron nodes x 11 , x 12 , x 21 , x 22 , …, x n1 and x n2 , which respectively correspond to the data of 2n sampled gases, including gas 1 concentration, gas 1 concentration change rate, gas 2 concentration, gas 2 concentration change rate, …, gas n-1 concentration, gas n-1 concentration change rate, micro-particle concentration and micro-particle concentration change rate; wherein n represents a natural number; the gas 1 concentration change rate, gas 2 concentration change rate, …, gas n-1 concentration change rate and micro-particle concentration change rate are the concentration change values of the gas or micro-particle per minute; The function iteration layer is set with 3 neuron nodes, respectively corresponding to a first function f(u) = 1 / (1+e u ), a second function f(u) = (e u -e -u ) / (e u +e -u ) and a third function f(u) = max(0, u); The early warning level output layer is provided with one neuron node corresponding to the fourth function f(x)=(e x - e -x ) / (e x + e -x ). an estimation optimization module for weighting and averaging the BP network estimation result and the curve estimation result according to a set weight ratio, and obtaining a final estimation result of the current safety condition of the metal high-temperature equipment, wherein the BP network estimation result accounts for 60% of the weight, and the curve estimation result accounts for 40% of the weight; an early warning output module for determining a fire hazard warning level according to the final estimation result of the current safety condition of the metal high-temperature equipment and outputting.

2. The metal material electrical current overload hazard tracing system of claim 1, wherein: The marking material is polyvinyl chloride, polyamide fiber, chlorobutyl rubber, polyurethane or polyvinyl chloride amide; and the sampled gas is H2, CO, HCL, CO2, VOCs and various hydrocarbon gases.

3. The metal material electrical current overload hazard tracing system of claim 1, wherein: The particle size of the micro-particles is between 2 nm and 20 μm.

4. The metal material electrical current overload hazard tracing system of claim 1, wherein: The application place of the system is a wind turbine cabin, a switch cabinet or an energy storage cabin.

5. A method of tracing a metal material current overload hazard, characterized by, The system comprises the following steps: A. coating a marking material on the surface of a specified position of a metal high-temperature equipment and numbering; B. obtaining the number of the marking material, and calling the thermal decomposition curve corresponding to the number of the marking material; C. sampling gas in the surrounding space of the specified position of the metal high-temperature equipment; D. extracting and analyzing the data of the sampled micro-particles and characteristic gas, matching the thermal decomposition curve of the marking material, and obtaining a curve estimation result of the current safety condition of the metal high-temperature equipment; E. inputting the data of the sampled gas into a trained BP neural network model, and the specific steps for training the BP neural network model are as follows: E11. preparing and arranging an input vector X and a target vector P for training the BP neural network model; the input vector X is a data set composed of the concentration of gas 1, the concentration change rate of gas 1, the concentration of gas 2, the concentration change rate of gas 2, …, the concentration of gas n-1, the concentration change rate of gas n-1, the concentration of micro-particles and the concentration change rate of micro-particles; and the target vector P is the fire hazard warning level corresponding to each parameter range; E12. substituting the input vector X into the BP neural network model, and iteratively calculating with the known parameters trained; E13. calculating the error between the output y and the target vector P; E14, judging whether the error falls between 0.01-0.1, if yes, stopping training; otherwise, adjusting the value of the known parameter, and returning to step E12 for repeated iteration update; The current safety condition of the metal high-temperature equipment is estimated by using the BP neural network intelligent analysis algorithm as the BP network estimation result; The BP neural network intelligent analysis algorithm adopts a three-layer BP neural network model, and includes the following specific steps: E21. Set the sampling parameter input layer parameters x = [x 11 , x 12 , x 21 , x 22 ,..., x n1 , and x n2 ]; E22、the sampling parameter input layer parameter x is input into a function iteration layer composed of a first function f(u)=1 / (1+e u ), a second function f(u)=(e u -e -u ) / (e u +e -u ) and a third function f(u)=max(0,u) to calculate with the known parameters generated by automatic iteration in the trained model; E23, input the calculation result of step E22 into the early warning level output layer composed of the fourth function f(x) = (e x - e -x ) / (e x + e -x ) to obtain the value of the output y = f(x); E24, determining the fire hazard warning level according to the calculation result of step E23, and determining the fire hazard warning level according to the value of y=f(x) falling in the range: when y∈[0, 0.15), it represents safety; when y∈[0.15, 0.3), it represents a first-level warning; When y∈[0.3, 0.45), it represents a second-level warning; When y∈[0.45, 0.6), it represents a third-level warning; When y∈[0.6, 0.75), it represents a fourth-level warning; When y∈[0.75, 1), it represents a fifth-level warning; F, the BP network estimation result and the curve estimation result are weighted and averaged according to the set weight ratio to obtain the final estimation result of the current safety condition of the metal high-temperature equipment, wherein the BP network estimation result accounts for 60% of the weight, and the curve estimation result accounts for 40% of the weight; G, determining the warning level of the fire hazard according to the final estimation result of the current safety condition of the metal high-temperature equipment and outputting.

6. The method of claim 5, wherein: The marking material is polyvinyl chloride, polyamide fiber, chlorobutyl rubber, polyurethane or polyvinyl chloride amide; the sampled gas is H2, CO, HCL, CO2, VOCs and various hydrocarbon gases.

7. The method of claim 5, wherein: The particle size of the microparticles is between 2nm and 20um.

8. The method of claim 5, wherein: The application site of the method is a wind turbine cabin, a switch cabinet or an energy storage cabin.

Images