Power cable fire early warning method based on gas sensing

Abstract

The invention discloses a power cable fire early warning method based on gas sensing, and the method comprises the steps: selecting different power cables to build a power cable overheating fault simulation device, and collecting gas released after the power cables are overheated; sending the gas into a gas chromatograph-mass spectrometer to obtain chromatograms of component compounds so as to identify the components of the component compounds, calculating intra-cluster errors of different cluster numbers based on the components of the component compounds by using an unsupervised K-Means clustering algorithm, and obtaining a tricluster result so as to select the characteristic gas of the cable fire, according to the characteristic gas, MOS gas sensors are selected to construct a gas sensor array, and dynamic response signals generated when the cable is heated to different temperatures within a certain time are used as original data for simulating fire accidents of the power cable. And based on the original data, optimizing the gas sensor array by using an algorithm of a long-short-term memory response sequence based on the characteristic quantity, and realizing early warning of the fire hazard of the power cable by adopting the optimized gas sensor array.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power cable fire early warning, and particularly relates to a power cable fire early warning method based on gas sensing. BACKGROUND

[0002] In recent years, due to short circuit, electric arc, external heat source, poor contact and other reasons, power cable fire accidents occur frequently, which causes serious consequences to power systems and economic life, and a fire alarm system should be set to avoid serious accidents. The national standard GB 50116-2013 stipulates that the protection object is a power supply line above 1000V, and a grating optical fiber temperature measuring type or infrared temperature measuring type electrical fire monitoring detector should be selected for the power cable. The grating optical fiber temperature measuring type electrical fire monitoring detector needs to be arranged in advance on the surface of the power cable and is expensive, and the infrared temperature measuring type electrical fire monitoring detector will have measurement errors due to human or smoke factors. In the early stage of fire, gas volatiles are produced earlier than smoke, and the common smoke type electrical fire monitoring detector often misses the best period of power cable fire warning when it alarms.

[0003] According to the detection principle, the power equipment fire warning gas sensor is divided into semiconductor type, catalytic combustion type, electrochemical type and infrared type. The metal oxide semiconductor (MOS) gas sensor has the advantages of low cost, high sensitivity, fast response, simple preparation process and easy micro-nano integration. At a certain working temperature, the MOS sensor surface undergoes oxidation-reduction reaction, causing the overall resistance value to change, and then detecting the composition and concentration of the measured gas. However, the cross-sensitivity of the semiconductor sensor limits specific detection and will respond to gases other than the target gas, which is an unavoidable defect at present. The gas composition of early cable fire is very complex, such as aromatic compounds, esters, aldehydes, ketones, saturated hydrocarbons and unsaturated hydrocarbons, and the cross-sensitivity of the MOS gas sensor limits the accuracy of fire warning.

[0004] The above information disclosed in the background section is only used to enhance the understanding of the background of the present application, and therefore can contain information that is not prior art known to those of ordinary skill in the art. SUMMARY

[0005] In view of the deficiencies in the prior art, the present application aims to provide a power cable fire early warning method based on gas sensing, which realizes more accurate power cable fire warning in the early stage of fire.

[0006] In order to achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] The application discloses a power cable fire early warning method based on gas sensing.

[0008] Different power cables are selected to build a power cable overheating fault simulation device to collect the gas released after the power cable overheats.

[0009] The gas is sent into a gas chromatograph mass spectrometer to obtain a chromatogram of component compounds, so that the component compound composition is identified, and the change of the absorption peak of the power cable outer sheath before and after being heated is obtained by an infrared spectrometer,

[0010] The K-Means clustering algorithm is used to calculate the intra-cluster error of different cluster numbers based on the component compound composition, and a three-cluster result is obtained to select the characteristic gas of the cable fire.

[0011] According to the characteristic gas, a MOS gas sensor is selected to build a gas sensor array, the MOS gas sensor is arranged in a ring shape and uniformly, a test circuit is built, and the output voltage and the circuit voltage satisfy the following formula: V C is the circuit voltage, R H is a heating resistor, R S is an unknown sensor resistor, R L is a known load resistor, the amplitude of the output voltage on the load resistor increases, and the increase or decrease of the output directly reflects the change of the concentration, and then reflects the environmental change caused by the high temperature of the cable, the response time, the recovery time, the response value of each MOS gas sensor in the steady-state gas environment and the dynamic response curve in the dynamic gas environment are tested,

[0012] The dynamic response signal of the cable heated to different temperatures within a certain time is used as original data simulating the occurrence of a power cable fire accident.

[0013] Based on the original data, a long short-term memory response sequence algorithm based on characteristic quantities is used to optimize the gas sensor array, and the optimized gas sensor array is used to warn the power cable fire.

[0014] In the method, the power cable overheating fault simulation device comprises a heater for heating the power cable and a gas bag for collecting the gas.

[0015] In the method, the characteristic gas comprises acetophenone, isooctanol and benzene.

[0016] In the method, the MOS gas sensor in the gas sensor array is placed directly above the power cable and is spaced apart from the power cable by a predetermined distance.

[0017] In the method, the original data is filtered and then input into the long short-term memory response sequence algorithm based on characteristic quantities for optimization training.

[0018] In the method, in the algorithm of the long short-term memory response sequence based on the characteristic quantity, a threshold value is set for fire warning by using a response value or a response change rate, and the output of a previous node to the input of a next node in the LSTM unit structure contains complete vectors, the LSTM unit removes or adds information on the information transmission path through a "gate", the "gate" is composed of a sigmoid function and a dot multiplication operation, the LSTM unit includes a forgetting gate that accepts the output of a previous unit and decides which information of Ct-1 to retain and discard, a memory gate that determines what kind of new information is stored in the cell state, and an output gate that determines the output value according to the cell state, 0 indicating complete forgetting of the information, and 1 indicating complete memory.

[0019] Advantages

[0020] The application selects a gas sensor array applied to early warning of cable fire based on the steady-state response curve and the dynamic response curve of the sensor. Based on the response characteristic data of the sensor, the early stage of cable fire is identified by using a pattern recognition method, and the sensor array is further optimized, so that more accurate power cable fire early warning in the early stage of fire can be realized.

[0021] The above description is only a summary of the technical solutions of the application. In order to make the technical means of the application more clear and understandable, and to enable those skilled in the art to implement the content of the description, and to make the above and other purposes, characteristics and advantages of the application more obvious and easy to understand, the following will be illustrated by specific embodiments of the application. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments or prior art of the present application, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings.

[0023] Through reading the detailed description in the preferred specific embodiments below, various other advantages and benefits of the present application will become clear to those skilled in the art. The drawings in the specification are only for the purpose of illustrating the preferred embodiments, and are not considered as limiting the present application. Obviously, the drawings described below are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without paying creative labor. Moreover, the same reference numerals are used to represent the same parts throughout the drawings.

[0024] In the drawings:

[0025] Figure 1 is a flowchart of a gas sensing based power cable fire early warning method according to an embodiment of the present disclosure;

[0026] Fig. 2(a), Fig. 2(b) are gas chromatogram results of a gas sensing based power cable fire early warning method according to an embodiment of the present disclosure, Fig. 2(a) is a gas chromatogram at 120℃, and Fig. 2(b) is a gas chromatogram at 180℃;

[0027] Figure 3 is a three cluster result diagram of a gas sensing based power cable fire early warning method according to an embodiment of the present disclosure;

[0028] Fig. 4(a), Fig. 4(b) are sensor basic test circuit diagrams of a gas sensing based power cable fire early warning method according to an embodiment of the present disclosure, Fig. 4(a) is a sensor basic test circuit, and Fig. 4(b) is a basic test circuit of TGS822;

[0029] Figure 5 is an LSTM unit structure diagram of a gas sensing based power cable fire early warning method according to an embodiment of the present disclosure.

[0030] The present application will be further explained in conjunction with the drawings and embodiments. DETAILED DESCRIPTION

[0031] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0032] Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0033] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0034] In order for those skilled in the art to better understand the technical solutions of the present application, the present application will be further described in detail below with reference to the drawings, and each of the drawings does not constitute a limitation on the embodiments of the present application.

[0035] In one embodiment, as shown in Figures 1 to 5 The present disclosure provides a gas sensing-based power cable fire early warning method, which comprises the following steps:

[0036] Different power cables are selected to build a power cable overheating fault simulation device to collect the gases released after the power cable overheats;

[0037] The gases are sent into a gas chromatograph-mass spectrometer to obtain a chromatogram of the component compounds to identify the component compound composition, and the infrared spectrometer is used to obtain the change in the absorption peak of the power cable outer sheath before and after heating,

[0038] The infrared spectrometer can obtain the gas chromatogram of the product at different temperatures, and the abscissa of the chromatogram is the retention time, and the ordinate is the relative abundance. Each peak represents a substance, and the peak area represents the content of the substance, as shown in FIG. 2(a) and FIG. 2(b). The gas production results of the cables at different temperatures are different, which is specifically manifested in the appearance time and peak area of the peaks, which confirms the statement that the gas products of the cable at high temperature carry temperature information.

[0039] Only a few gases that can cover most of the temperature information are selected as the characteristic gases of the early cable fire, which has more practical value. The content values of the gases at different temperatures are used to divide the gases into different categories. Each category represents a similar temperature information, and appropriate gases are selected from each category to obtain the characteristic gases that can represent the early stage of the cable fire.

[0040] The K-Means clustering algorithm is used to calculate the intra-cluster error of different cluster numbers based on the component compound composition, and three cluster results are obtained to select the characteristic gases of the cable fire, as shown in Figure 3 ,

[0041] The cosine similarity between two high-dimensional vectors is calculated to compare the similarity between the gases. First, two samples with the lowest similarity are selected as the initial centroids, and the error in each cluster after two classifications is calculated as an evaluation of the classification quality. Then, the sample farthest from the two initial centroids is selected as the third initial centroid, and the above calculation is repeated to obtain a new evaluation value, until the fourth initial centroid, the fifth initial centroid, etc. Finally, the error values at different classification numbers are compared, and the centroid number with the smallest error and the least classification number is selected. The intra-cluster error of different cluster numbers is calculated, and one, two, three, four and five categories are divided. When all the gases are divided into three categories, the error has reached a very low level, and the downward trend tends to be flat.

[0042] Figure 3 The results of the three clusters are shown. Isooctanol alone belongs to a cluster. While 2-phenylisopropanol and isooctanol are both alcohols, the same hydroxyl group might make it difficult for the gas sensor to identify, and the carbonyl group of acetophenone is more easily detected, so acetophenone is more representative in the sense of detection. Benzene vapor only appears at high temperatures above 170℃, and the content is high, so it can be used as a warning of extremely dangerous gas and should also be selected.

[0043] According to the characteristic gas, the MOS gas sensor is selected to build a gas sensor array, and the MOS gas sensor is arranged uniformly in a ring shape. A test circuit is built, and 5V stable voltage is used as the circuit voltage and heating voltage, 4.7kΩ voltage dividing resistor is connected in series with the sensor, and the resistance change of the gas sensor is reflected in real time, as shown in FIG. 4(a) and FIG. 4(b).

[0044] In FIG. 4(a), V C is the circuit voltage, V H is the heating voltage, V OUT is the output voltage, R H is the heating resistor, R S is the unknown sensor resistance, R L is the known load resistance. The output voltage is measurable, and the circuit voltage satisfies equation (1):

[0045]

[0046] The gas sensor is placed at a high potential, and as the concentration of the gas to be measured increases, the resistance of the sensor decreases. At this time, the amplitude of the output voltage on the load resistor increases, and the increase and decrease of the output will directly reflect the change in concentration, and then reflect the environmental changes caused by high temperature of the cable. The test circuit tests the response time, recovery time, response value of each MOS gas sensor in the steady-state gas environment, and the dynamic response curve in the dynamic gas environment,

[0047] PVC cable can withstand 90℃ working temperature for a long time, so the initial state of dynamic environment is the gas environment formed by the long-term maintenance of cable outer sheath at 90℃, that is, the initial environment of dynamic test corresponds to the steady-state environment of cable outer skin at 90℃. Considering the characteristics of cable temperature rise leading to fire in a short time, it is set to rise to the preset temperature only for 10 min, and the dynamic response of the sensor is measured when the outer skin rises from 90℃ to 110℃, 130℃, 150℃ and 170℃, and the average temperature rise rate is 2℃ / min, 4℃ / min, 6℃ / min and 8℃ / min respectively. The resistance of the gas sensor in the gas atmosphere of the cable at 90℃ is Ri, and the resistance value Rg during the test, the dynamic response value S during the test is Ri / Rg, and the dynamic response curve in the dynamic gas environment can be obtained.

[0048] The dynamic response signal of the cable heated to different temperatures within a certain time is used as the original data of the simulation of power cable fire accident,

[0049] Based on the original data, the gas sensor array is optimized using the algorithm based on the long short-term memory response sequence of characteristic quantity, and the optimized gas sensor array is used for early warning of power cable fire.

[0050] The dynamic response curve of the gas sensor is essentially a numerical sequence sorted by time, and the responses at different times are related in some way, which can reflect the possible rules behind. Long short-term memory (LSTM) neural network model is a variant of recurrent neural network (RNN), which is good at processing time series data.

[0051] The specific steps are as follows:

[0052] 1) Environment is MATLAB. The curve with dynamic response greater than 1.05 is divided by 20s, each time sequence interval is 10s, and each sequence has 200 points. A 1*251 cell array is constructed, and an 8*200 multivariate response matrix exists in each cell. 8 is the number of sensors, and 200 is the number of time series under multiple temperature rise rates. Each time series has a corresponding category, so a 251*1 category matrix is also constructed. The training set and test set are roughly allocated according to the ratio of 4:1.

[0053] 2) The LSTM network is a seven-layer architecture, which is input layer, hidden layer, activation function layer, Dropout layer, fully connected layer, output layer and classification layer in turn. The input layer is 8, which is determined by the number of sensors. The hidden layer has a bidirectional LSTM structure, which contains 400 hidden units and outputs the last value in the time series. The activation function layer uses reluLayer, and the output is a linear combination of the outputs of the LSTM units at the last time, which combines the output features of each hidden unit. The Dropout layer can randomly delete some hidden layer nodes to avoid overfitting during training, and the dropout ratio is set to 10%. The fully connected layer is 2, which corresponds to the number of output categories. The output layer uses the softmax function to convert the output value of multi-classification into a regular probability distribution. The classification layer converts the data output by the softmax function into the corresponding category. The training options specify that the "adam" solver is used to calculate in the CPU environment, the maximum training rounds are 100, the small batch size is 200, and the data is not shuffled.

[0054] 3) The LSTM network is trained using the trainNetwork function with the above specified manner. The network is trained for 73 rounds using the data of the training set, and a high accuracy is achieved, and the loss becomes very small.

[0055] 4) The test set data is tested through the trained LSTM network using the classify function to test the classification accuracy, and the accuracy reaches 98.04%.

[0056] In a preferred embodiment of the method, the power cable overheating fault simulation device comprises a heater for heating the power cable and a gas bag for collecting gas.

[0057] In a preferred embodiment of the method, the characteristic gas comprises acetophenone, isooctanol and benzene.

[0058] In a preferred embodiment of the method, in the gas sensor array, the MOS gas sensor is placed directly above the power cable and spaced a predetermined distance from the power cable.

[0059] In a preferred embodiment of the method, the raw data is filtered and input into the algorithm optimization training based on the long short-term memory response sequence of the characteristic quantity.

[0060] In a preferred embodiment of the method, in the algorithm based on the long short-term memory response sequence of the characteristic quantity, a reasonable threshold value needs to be set for the response value or the response change rate to warn of fire, and the LSTM unit structure diagram is as shown in Figure 5

[0061] ​Structure: The LSTM unit structure mainly consists of three gates (forget gate, input gate, output gate) and a memory cell. Each gate contains a sigmoid activation function to control the degree of information passing. The memory cell is used to store long-term information. In LSTM, each unit is connected to the previous and next units.

[0062] Connection: Specifically, the input x_t at each time and the output h_t-1 at the previous time are taken as the input of the current LSTM unit, and the output h_t and the memory cell state c_t at the current time are taken as the input of the next LSTM unit. This connection allows LSTM to handle sequence data of arbitrary length.

[0063] Input path: At each time, the LSTM unit receives input x_t and output h_t-1 at the previous time. These inputs are first processed through the forget gate, input gate and output gate to produce corresponding coefficients and candidate states. Then, these coefficients and states are used to update the state of the memory cell and produce the output h_t at the current time.

[0064] Output path: The output h_t of the LSTM unit not only serves as the result at the current time, but also as one of the inputs at the next time. At the same time, the state of the memory cell c_t is also transmitted internally to maintain long-term memory.

[0065] Corresponding processing: In processing input and output, LSTM uses various nonlinear functions and gating mechanisms to ensure effective transmission and updating of information. Specifically, the sigmoid function is used to produce gating coefficients to control the degree of information passing; the tanh function is used to produce the activation values of candidate states and memory cell states; and the gating mechanism is used to achieve precise control of information through multiplication operations.

[0066] Figure 5 The output of the previous node to the input of the next node, which contains complete vectors. The multiplication sign represents dot operation, the plus sign represents vector addition or vector concatenation, and σ represents the sigmoid function, and tanh represents the hyperbolic tangent function. In LSTM, the cell state, which is the transmission path of information, is Figure 5The horizontal line in the middle of the top is the long-term memory line. Information flows along the transmission path, similar to the principle of a conveyor belt. There are only linear interactions on the transmission path, which maintains the flow of information while maintaining its stability to some extent. LSTM has the ability to remove or add information on the information transmission path, which is achieved through "gates," which are a way to select information to pass through, consisting of a sigmoid function and a dot product operation. There are three gate structures in LSTM, namely the forget gate, the memory gate, and the output gate. The forget gate can decide to discard some information, that is, accept the output of the previous unit and decide which information of Ct-1 to keep and discard, and finally get a value between 0 and 1, 0 means completely forgetting the information, and 1 means keeping the complete memory. The memory gate can determine what new information can be stored in the cell state. For the attribute information discarded in the forget gate, find the corresponding new attribute information at the memory gate to add it to make up for the discarded attribute information. The output gate determines the output value according to the cell state. In this way, LSTM can long-term memory important information, and the memory can be dynamically adjusted with the input.

[0067] In one embodiment, different materials and different models of power cables are selected to build a power cable overheating fault simulation device, and a gas bag is used to collect the gas released after the power cable overheats. The early cable fire gas is sent into a gas chromatograph mass spectrometer to obtain a chromatogram of the component compounds, as shown in FIGS. 2(a) and 2(b), and it is identified that the products include benzene, alcohol, ketone, olefin, alkane, and lipid. The changes in the absorption peaks of the cable sheath before and after heating are obtained by an infrared spectrometer to analyze the chemical changes of the cable sheath at high temperatures, thereby exploring the source of the early cable fire gas. Using unsupervised K-Means clustering algorithm, considering the original data of the early fire gas, the intra-cluster error of different cluster numbers is calculated, and the three-cluster result is as shown in FIG. 3(a) and FIG. 3(b). Selecting acetophenone, isooctanol, and benzene as the characteristic gases of the early cable fire, which can cover most of the temperature information. Figure 3

[0068] Based on the gas production results, a number of MOS gas sensors of different models are selected to obtain a gas sensor array, and each sensor is arranged in a ring shape. The test circuit is built as shown in FIGS. 4(a) and 4(b), V C is the circuit voltage, V H is the heating voltage, V OUT is the output voltage, R H is the heating resistance, R S is the unknown sensor resistance, R L is the known load resistance. The output voltage and the circuit voltage satisfy:

[0069] ​

[0070] The resistance change of the gas sensor is converted into a voltage change that is easier to collect. The gas sensor is placed at a high potential, and as the concentration of the gas to be measured increases, the resistance of the sensor decreases continuously. At this time, the amplitude of the output voltage on the load resistance increases, and the increase or decrease of the output will directly reflect the change in concentration, and further reflect the environmental changes caused by high temperature of the cable. The response time, recovery time, response value of each sensor in a stable gas environment, and the dynamic response curve in a dynamic gas environment are tested, and the gas sensor that can be applied to the early warning of cable fire is selected, that is, the MOS gas sensor with fast response time, short recovery time, high response value and strong dynamic response, such as TGS 822, TGS2600, TGS2602, TGS2603, TGS2609, TGS2610, TGS2620, MQ3B and eight other sensors. The response characteristics of the sensor under different placement positions are tested, and it is found that the response time and final response strength of the gas sensor placed directly above the cable are stronger than those placed directly below. However, the gas sensor should not be too close to the cable, because there is a risk of insulation breakdown, and because too high gas concentration will cause the sensor to respond strongly and cause the sensor to malfunction.

[0071] The dynamic response signal of the cable heated to different temperatures within a certain time is used as the original data simulating the occurrence of power cable fire accidents. After filtering to improve the quality of the original data, the early warning method of power cable fire based on characteristic quantity and the algorithm of long short-term memory (LSTM) response sequence are compared. The early warning method of power cable fire based on characteristic quantity is mutually restricted in terms of warning accuracy and warning speed. The use of response value or response change rate to warn fire requires setting a reasonable threshold value. The structure diagram of the LSTM unit is as shown in Figure 5 Figure 5 The output of the upper node is input to the next node, and both contain complete vectors. The multiplication sign represents dot operation, the plus sign represents vector addition or vector splicing, and sigma represents sigmoid function and tanh represents hyperbolic tangent function. In LSTM, the cell state, that is, the information transmission path, is important, that is, Figure 5 ​The middle top horizontal line is the long-term memory line. Information flows on the transmission path, which is similar to the principle of a conveyor belt. There are only some linear interactions on the transmission path, so that the flow of information can be maintained, and the stability of the information can be maintained to some extent. LSTM has the ability to remove or add information on the information transmission path, and this ability is achieved through "gates", which are a way of selecting information to pass through, consisting of a sigmoid function and a dot product operation. There are three gate structures in LSTM, namely the forget gate, the memory gate and the output gate. The forget gate can decide to discard some information, that is, to accept the output of the previous unit and decide which information to keep and discard C t-1 , and finally get a value between 0 and 1, 0 means completely forgetting the information, and 1 means keeping the complete memory. The memory gate can determine what new information can be stored in the cell state. For the attribute information discarded in the forget gate, find the corresponding new attribute information at the memory gate and add it to make up for the discarded attribute information. The output gate determines the output value according to the cell state. In this way, LSTM can long-term memory important information, and the memory can be dynamically adjusted with the input.

[0072] The LSTM neural network does not have to consider specific features and can directly use dynamic responses to train the network for time series, improving the ability to identify cable fires and the accuracy of identification. According to the similarity of the response, the redundant sensor MQ3B is removed, and the number and array of sensors are optimized. Based on the high-temperature gas production of the cable, the MOS gas sensor array responds to a certain concentration of characteristic gas, transmits the fire fault information to the upper computer to realize early warning of power cable fire.

[0073] Although the embodiments of the present application are described above in combination with the drawings, the present application is not limited to the above specific embodiments and application fields, and the above specific embodiments are only illustrative and guiding, but not limiting. Those skilled in the art can make many forms under the guidance of this specification and without departing from the scope protected by the claims of the present application, which are all within the protection of the present application.

Claims

1. A method for early warning of power cable fires based on gas sensing, characterized in that, It includes the following steps: Different power cables were selected to build a power cable overheating fault simulation device to collect the gas released after the power cable overheated; The gas was fed into a gas chromatograph-mass spectrometer to obtain chromatograms of the component compounds, thereby identifying the components. Infrared spectroscopy was used to obtain changes in the absorption peaks of the power cable's outer sheath before and after heating. The unsupervised K-Means clustering algorithm was used to calculate the intra-cluster error based on the component compound composition, and the tri-cluster results were obtained to select the characteristic gases of cable fires. Based on the characteristic gas, select MOS gas sensors to construct a gas sensor array. The MOS gas sensors are arranged in a uniform ring. Build a test circuit, and the output voltage and circuit voltage satisfy the following: V C It is the circuit voltage, R H It is a heating resistor, R S The sensor resistance, R, is unknown. L The load resistance is known. The dynamic response signals of the cable heated to different temperatures within a certain period of time are used as the raw data for simulating power cable fire accidents. Based on the original data, an algorithm based on long short-term memory response sequences of feature quantities is used to optimize the gas sensor array, and the optimized gas sensor array is used to provide early warning of power cable fires.

2. The method according to claim 1, characterized in that, Preferably, the power cable overheating fault simulation device includes a heater for heating the power cable and a gas bag for collecting gas, using the gas bag to collect the gas released after the power cable overheats.

3. The method according to claim 1, characterized in that, Characteristic gases include acetophenone, isooctyl alcohol, and benzene.

4. The method according to claim 1, characterized in that, In the gas sensor array, the MOS gas sensor is placed directly above the power cable.

5. The method according to claim 1, characterized in that, The original data is filtered and then input into an algorithm optimization training based on the long short-term memory response sequence of the feature quantity.

6. The method according to claim 1, characterized in that, In algorithms based on long short-term memory response sequences of feature quantities, threshold values ​​are set for fire early warning using response values ​​or response change rates.

Images