A sensor array based method for detecting overheat in electrical equipment

Abstract

The application provides a kind of electrical equipment overheating detection method based on sensor array.The method comprises: obtaining the output data of sensor in sensor array, the sensor is used to measure the concentration of different volatile gas generated in electrical equipment working environment;The output data of sensor is input into the detection model based on BP neural network, and the electrical equipment overheating detection result is obtained: not overheating, epoxy resin overheating, PVC overheating, ABS overheating;When the detection result is epoxy resin overheating or PVC overheating or ABS overheating, and the duration exceeds the set threshold, an alarm signal is sent.The application can effectively detect whether the electrical equipment is overheated by detecting the concentration of gas volatiles generated when the electrical equipment is heated, without setting temperature sensors at each monitoring point, and can identify the material category of overheating, so as to quickly locate the fault source based on the overheating material category, and reduce the hardware cost.

Description

Technical Field

[0001] This invention belongs to the field of electrical equipment safety testing technology, specifically relating to an overheating detection method for electrical equipment based on a sensor array. Background Technology

[0002] Early warning systems for electrical fires are crucial for their prevention. Currently, the common approach is to install temperature sensors at multiple monitoring points on electrical equipment to detect overheating. However, many components within electrical equipment are prone to overheating. While temperature detection can provide early warning, it often only detects localized overheating. To detect overheating in all areas of the equipment would require installing a large number of temperature sensors, resulting in significant costs.

[0003] In view of this, the present invention proposes an overheating detection method for electrical equipment based on a sensor array. This method detects whether the electrical equipment is overheating by detecting the concentration of volatile gases generated when the equipment heats up, and identifies the type of material being overheated. It can be used for early warning of electrical fires. Summary of the Invention

[0004] To address the aforementioned problems in the prior art, this invention provides a method for detecting overheating in electrical equipment based on a sensor array.

[0005] To achieve the above objectives, the present invention adopts the following technical solution.

[0006] A method for detecting overheating in electrical equipment based on a sensor array includes the following steps:

[0007] Acquire output data from sensors in a sensor array, wherein the sensors are used to measure the concentrations of different volatile gases generated in the working environment of electrical equipment;

[0008] The sensor's output data is input into a detection model based on a BP neural network to obtain the electrical equipment overheating detection results: no overheating, epoxy resin overheating, PVC overheating, ABS overheating;

[0009] An alarm signal is issued when the detection result indicates that the epoxy resin, PVC, or ABS is overheated, and the duration exceeds the set threshold.

[0010] Furthermore, the sensor includes:

[0011] The first sensor is used to detect hydroxyl-containing compounds in volatile gases that have properties similar to alcohol;

[0012] The second sensor is used to detect flammable components in volatile gases that are similar in properties to methane and hydrogen.

[0013] The third sensor is used to detect volatile organic compounds in gaseous volatiles that are similar in properties to benzene and phenol;

[0014] The fourth sensor is used to detect methyl-containing components in volatile gases that have properties similar to methane;

[0015] The fifth sensor is used to detect methyl-containing components in volatile gases that have properties similar to ethane;

[0016] The sixth sensor is used to detect small molecule components in volatile gases that are similar in properties to natural gas.

[0017] Furthermore, the first to sixth sensors are TGS822, TGS813, TGS2602, TGS2620, TGS2610 and TGS2611, respectively.

[0018] Furthermore, the sensor is a metal oxide gas sensor composed of gas-sensitive elements. The resistance of the gas-sensitive elements changes with the concentration of the detected gas; the higher the concentration of the detected gas, the lower the resistance of the gas-sensitive elements. The output voltage of the sensor is:

[0019]

[0020] In the formula, V is the output voltage of the sensor, and K S R is the signal amplification factor. S R0 is the resistance of the gas-sensitive element, C is the reference resistance, and C is an inherent constant.

[0021] Furthermore, the method for training the detection model includes:

[0022] Data set S was obtained through simulation experiments. The data in S consisted of sensor output data at different temperatures when different types of materials were used to simulate the different volatile gases. The types of materials included epoxy resin, PVC, and ABS.

[0023] Each material has an overheating temperature threshold. By comparing the temperature corresponding to each data point in S with the overheating temperature threshold of the corresponding material, a detection result label is obtained for each data point. The detection result label includes no overheating, epoxy resin overheating, PVC overheating, and ABS overheating.

[0024] Each data point in S is labeled based on the obtained detection result labels to obtain a dataset S0 including the detection result labels;

[0025] Half of the sample data is extracted from S0 to obtain the training dataset S1, and the other half of the sample data forms the test dataset S2;

[0026] The detection model is trained using the training dataset S1, and tested using the test dataset S2.

[0027] Furthermore, the method for obtaining the dataset S through simulation experiments includes:

[0028] S1. A simulation experimental device that connects a sensor array, a data acquisition module, a host computer, and a heating stage, wherein the sensor array and the heating stage are housed in the same cavity;

[0029] S2. Set the temperature of the heating platform. After the temperature stabilizes, read the output data of each sensor in the sensor array to obtain the initial value of the sensor. The initial value of the i-th sensor is represented as x. 0i Where i = 1, 2, ..., 6;

[0030] S3. Place a type of material on the tray of the heating table, heat it at a constant temperature for a period of time, and read n output data from each sensor according to the acquisition period T. The output data of the i-th sensor is x'. i Calculate x i =x' i -x 0i The categories of materials include epoxy resin, PVC, and ABS.

[0031] S4. Release the remaining gas in the cavity, then proceed to step S3 and replace the material on the tray with another material; continue until all three materials have been replaced.

[0032] S5. Increase the heating stage temperature by a step size, repeating steps S2 to S5 until the heating stage temperature reaches the set upper temperature limit, obtaining the dataset S represented in matrix form:

[0033]

[0034] In the formula, N = 3 × m × n, where m is the number of temperature points set on the heating platform.

[0035] Furthermore, the method also includes: first normalizing each data point in S0, and then dividing the normalized S0 into S1 and S2; the normalization formula is:

[0036]

[0037] In the formula, x i - For any data x in the i-th column of S0 i The normalized value, x i-min x i-max Let i and y be the minimum and maximum values ​​of the i-th column of S0, respectively, where i = 1, 2, ..., 6.

[0038] Furthermore, the method also includes: performing dimensionality reduction processing on the training dataset S1 and the test dataset S2, and training and testing the detection model based on the dimensionality-reduced training dataset and test dataset.

[0039] Furthermore, methods for dimensionality reduction of the training dataset S1 include:

[0040] Running the PCA algorithm on S1 yields a linear transformation matrix W1 with 1 rows and 6 columns, where w1 = 1;

[0041] Running the LDA algorithm on S1 yields a linear transformation matrix W2 with 6 columns and 2 rows, where 1 ≤ w2 ≤ 3;

[0042] Combine W1 and W2 into a matrix W with 6 rows and 6 columns, consisting of w1+w2 rows.

[0043] Dimensionality reduction of S1 is performed using the following formula:

[0044] S 1- =W×(S1) T

[0045] In the formula, S 1- This is the matrix after S1 has been reduced in dimension.

[0046] Furthermore, the detection model is a second-order BP neural network consisting of a material identification network and an overheating diagnosis network; the material identification network is used to determine the category of the heated material, and the overheating diagnosis network is used to output the overheating detection result.

[0047] Compared with the prior art, the present invention has the following beneficial effects.

[0048] This invention acquires the output data from sensors in a sensor array and inputs this data into a detection model based on a backpropagation neural network. This yields overheating detection results for electrical equipment, including overheating of epoxy resin, PVC, and ABS. When the detection result indicates overheating of epoxy resin, PVC, or ABS, and the duration exceeds a set threshold, an alarm signal is issued, thus achieving overheating detection and alarm for electrical equipment. Furthermore, by detecting the concentration of volatile gases generated when electrical equipment heats up, this invention can effectively detect overheating without requiring temperature sensors at every monitoring point. It can also identify the type of overheated material, facilitating rapid fault location based on the overheated material type and reducing hardware costs. Attached Figure Description

[0049] Figure 1 This is a flowchart of an electrical equipment overheat detection method based on a sensor array, according to an embodiment of the present invention.

[0050] Figure 2 This is a schematic diagram of the simulation experimental setup. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer and more understandable, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0052] Figure 1 This is a flowchart of an overheat detection method for electrical equipment based on a sensor array, according to an embodiment of the present invention, including the following steps:

[0053] Step 101: Obtain the output data of the sensors in the sensor array, wherein the sensors are used to measure the concentration of different volatile gases generated in the working environment of the electrical equipment;

[0054] Step 102: Input the sensor output data into the detection model based on BP neural network to obtain the electrical equipment overheat detection results: no overheating, epoxy resin overheating, PVC overheating, ABS overheating;

[0055] Step 103: When the detection result indicates that the epoxy resin, PVC, or ABS is overheated, and the duration exceeds the set threshold, an alarm signal is issued.

[0056] This embodiment provides a method for detecting overheating of electrical equipment based on a sensor array. The hardware involved includes a sensor array placed in the environment near the monitored electrical equipment, a data acquisition unit connected to the sensor array, and a host computer connected to the data acquisition unit. The host computer can be directly connected to the data acquisition unit or configured as a cloud server communicating with the data acquisition unit via a communication module. The data acquisition unit collects gas concentration data output from each sensor in real time. The host computer identifies whether the electrical equipment is overheating based on the gas concentration data. The method described in this embodiment is executed by the host computer.

[0057] In this embodiment, step 101 is mainly used to acquire the output data of the sensors in the sensor array. Electrical equipment generates heat during operation, producing various volatile gases in its working environment. Generally, the higher the temperature of the electrical equipment, the higher the concentration of volatile gases in the environment. This embodiment is based on this characteristic, determining whether there is overheating of the electrical equipment by detecting the concentration of volatile gases in the environment. Since the volatile gases are a mixture of various components, this embodiment uses a sensor array composed of multiple gas-sensitive sensors, with different sensors used to detect the concentration of different gas components. It is worth noting that since the purpose of this embodiment is not to measure the gas concentration, the voltage value output by the sensor is not necessarily equal to the gas concentration value, but rather a voltage value positively correlated with the gas concentration, generally referred to as the response value.

[0058] In this embodiment, step 102 is mainly used for overheat detection of electrical equipment. This embodiment constructs a detection model that takes the sensor's output data as input and the detection result as output. The real-time output data of the sensor is input into the trained detection model to obtain the detection result. The detection result in this embodiment includes not only whether overheating occurs, but also the type of material that is overheating, specifically including no overheating, epoxy resin overheating, PVC overheating, and ABS overheating. PVC, ABS, and epoxy resin are widely used in the non-metallic parts of electrical equipment. The volatile gases emitted by electrical equipment are mainly generated by the overheating of these three materials. PVC is a major component of cables and is often used for insulation and protection of electrical equipment; ABS is a material for electromagnetic shielding shells, terminals, and other components in electrical equipment and is used in various anti-static components; epoxy resin is a major material for printed circuit boards. Therefore, this embodiment defines the overheating material categories as PVC, ABS, and epoxy resin.

[0059] In this embodiment, step 103 is mainly used to issue an alarm based on the detection results. As mentioned above, there are four types of detection results: no overheating, epoxy resin overheating, PVC overheating, and ABS overheating. No overheating indicates that the electrical equipment is working normally and no alarm is needed. When the detection result is overheating of any material, the duration of the overheating detection result is further examined. If the duration does not exceed a set threshold, for example, if the detection result quickly changes back to no overheating, no alarm is needed; conversely, if the duration exceeds the set threshold, prolonged overheating may cause a fire, requiring an immediate alarm to prompt personnel to take relevant measures.

[0060] This embodiment uses a sensor array to detect the concentration of volatile gases in the working environment of electrical equipment, thereby identifying whether the electrical equipment is overheating. This eliminates the need to install temperature sensors at every monitoring point to effectively detect overheating of electrical equipment and can identify the type of overheated material. This facilitates rapid location of the fault source based on the type of overheated material, reducing hardware costs.

[0061] As an optional embodiment, the sensor includes:

[0062] The first sensor is used to detect hydroxyl-containing compounds in volatile gases that have properties similar to alcohol;

[0063] The second sensor is used to detect flammable components in volatile gases that are similar in properties to methane and hydrogen.

[0064] The third sensor is used to detect volatile organic compounds in gaseous volatiles that are similar in properties to benzene and phenol;

[0065] The fourth sensor is used to detect methyl-containing components in volatile gases that have properties similar to methane;

[0066] The fifth sensor is used to detect methyl-containing components in volatile gases that have properties similar to ethane;

[0067] The sixth sensor is used to detect small molecule components in volatile gases that are similar in properties to natural gas.

[0068] This embodiment presents a technical solution for a sensor array. The sensor array consists of six sensors, namely the first to the sixth sensors. Each of the six sensors is used to detect the concentration of volatile gases of different components. Because the composition of volatile gases produced during the operation of electrical equipment is complex and diverse, it is almost impossible to determine the presence of a certain volatile gas using a single gas sensor. To improve detection accuracy, it is necessary to set up multiple different sensors to detect volatile gases. Since the same sensor can respond to multiple gas components, relying on only a single sensor to detect a certain volatile gas component can easily lead to false alarms. Therefore, when designing the sensor array, it is necessary to ensure that the detection ranges of different gas sensors overlap to further improve the detection accuracy of the sensor array. Although there are currently no gas sensors specifically designed to detect volatile gases released from overheated electrical equipment, the functional groups contained in these volatile gases are very common, such as hydroxyl, methyl, ethyl, and benzene rings. However, gases containing these functional groups, such as ethanol, methane, ethane, and benzene, are gases that many existing gas sensors can detect. Considering that organic compounds containing the same functional groups have similar chemical properties, and that metal oxide gas sensors also produce similar outputs when detecting gases containing the same functional groups, it is feasible to select a sensor that detects gases with similar functional groups to the components of gaseous volatiles for the detection of gaseous volatiles.

[0069] As an optional embodiment, the first to sixth sensors are TGS822, TGS813, TGS2602, TGS2620, TGS2610 and TGS2611, respectively.

[0070] This embodiment provides one specific model of each of the six sensors. When performing overheat diagnosis of electrical equipment, the concentration of the detected gas is low while the ambient temperature is high; therefore, metal oxide gas sensors with advantages such as high sensitivity and high temperature resistance should be selected. The six sensors selected in this embodiment are TGS822, TGS813, TGS2602, TGS2620, TGS2610, and TGS2611, all of which are high-sensitivity, low-cost metal oxide gas sensors.

[0071] As an optional embodiment, the sensor is a metal oxide gas sensor composed of a gas-sensitive element. The resistance of the gas-sensitive element changes with the concentration of the detected gas; the higher the concentration of the detected gas, the lower the resistance of the gas-sensitive element. The output voltage of the sensor is:

[0072]

[0073] In the formula, V is the output voltage of the sensor, and K S R is the signal amplification factor. SR0 is the resistance of the gas-sensitive element, C is the reference resistance, and C is an inherent constant.

[0074] This embodiment provides an expression for the sensor's output voltage. The sensor in this embodiment mainly consists of a gas-sensitive element; the higher the gas concentration in the environment, the lower the resistance of the gas-sensitive element. The expression for the sensor's output voltage V is as shown above. Where, R... S R is the resistance of the gas-sensitive element; the higher the gas concentration, the lower the resistance. S The smaller the value, the larger V becomes. Therefore, the sensor output voltage V is positively correlated with the gas concentration.

[0075] As an optional embodiment, the method for training the detection model includes:

[0076] Data set S was obtained through simulation experiments. The data in S consisted of sensor output data at different temperatures when different types of materials were used to simulate the different volatile gases. The types of materials included epoxy resin, PVC, and ABS.

[0077] Each material has an overheating temperature threshold. By comparing the temperature corresponding to each data point in S with the overheating temperature threshold of the corresponding material, a detection result label is obtained for each data point. The detection result label includes no overheating, epoxy resin overheating, PVC overheating, and ABS overheating.

[0078] Each data point in S is labeled based on the obtained detection result labels to obtain a dataset S0 including the detection result labels;

[0079] Half of the sample data is extracted from S0 to obtain the training dataset S1, and the other half of the sample data forms the test dataset S2;

[0080] The detection model is trained using the training dataset S1, and tested using the test dataset S2.

[0081] This embodiment presents a technical solution for training a detection model. Supervised training of the model requires the establishment of a labeled training dataset. Therefore, the key to this embodiment is establishing the training dataset. Existing technologies generally use historical data to establish training datasets. Since the technical solution of this embodiment is proposed for the first time and there is no available historical data, this embodiment obtains the training dataset through simulation experiments. First, the output data of the sensor is obtained through simulation experiments at different temperatures using epoxy resin, PVC, and ABS to simulate different volatile gases (generated by electrical equipment), resulting in dataset S. Then, dataset S is labeled, that is, the detection result label of each data is labeled, including no overheating, epoxy resin overheating, PVC overheating, and ABS overheating. The labeling method is as follows: first, the overheating temperature thresholds of the three materials are set respectively, and then the temperature corresponding to each data is compared with the overheating temperature threshold of the corresponding material. If it exceeds the threshold, the corresponding material is overheated (e.g., ABS overheating); otherwise, it is not overheated. When the temperature is below 400℃, none of the three materials, PVC, ABS, and epoxy resin, can burn. As long as overheating detection is achieved before the temperature of these three materials reaches 400℃, early warning of electrical fires can be realized. Therefore, the overheating temperatures of ABS, epoxy resin, and PVC can be set to 200℃, 200℃, and 180℃, respectively. After annotation, dataset S0 is obtained, which is then divided into training dataset S1 and test dataset S2.

[0082] As an optional embodiment, the method for obtaining the dataset S through simulation experiments includes:

[0083] S1. A simulation experimental device that connects a sensor array, a data acquisition module, a host computer, and a heating stage, wherein the sensor array and the heating stage are housed in the same cavity;

[0084] S2. Set the temperature of the heating platform. After the temperature stabilizes, read the output data of each sensor in the sensor array to obtain the initial value of the sensor. The initial value of the i-th sensor is represented as x. 0i Where i = 1, 2, ..., 6;

[0085] S3. Place a type of material on the tray of the heating table, heat it at a constant temperature for a period of time, and read n output data from each sensor according to the acquisition period T. The output data of the i-th sensor is x'. i Calculate x i =x' i -x 0i The categories of materials include epoxy resin, PVC, and ABS.

[0086] S4. Release the remaining gas in the cavity, then proceed to step S3 and replace the material on the tray with another material; continue until all three materials have been replaced.

[0087] S5. Increase the heating stage temperature by a step size, and repeat steps S2 to S5 until the heating stage temperature reaches the set upper temperature limit, resulting in a dataset S represented in the following matrix form:

[0088]

[0089] In the formula, N = 3 × m × n, where m is the number of temperature points set on the heating platform.

[0090] This embodiment presents a technical solution for obtaining dataset S through simulation experiments. This embodiment includes steps S1 to S5.

[0091] Step S1 is mainly used to connect the simulation experiment device. The simulation experiment device in this embodiment is as follows: Figure 2 As shown, the system includes a sensor array, a data acquisition module, a host computer, and a heating platform. The sensor array and the heating platform are housed within the same cavity, the internal space of which simulates the working environment of electrical equipment. A tray for holding materials (epoxy resin, PVC, and ABS) is mounted on the heating platform, which maintains a constant temperature for the materials within the tray according to a set temperature.

[0092] It is worth noting that, in order to improve the accuracy of the experiment, all sensors must be placed in clean air and preheated for one day before the experiment, and the heating stage must be calibrated.

[0093] Step S2 is mainly used to obtain the initial value of the sensor when there is no volatile gas in the cavity. Ideally, a gas sensor should output 0 when the gas concentration is 0, but in reality, the sensor output is not 0, but has an initial value. In order for the sensor output to accurately reflect the gas concentration, the influence of the initial value should be eliminated, so it is necessary to measure the initial value first. In this embodiment, the initial value is obtained by measuring the sensor output value when no material is placed in the tray. During the experiment, the heating stage temperature is first set, and after the heating stage temperature stabilizes and the output voltage of each sensor stabilizes, the output of each sensor is recorded as the initial value.

[0094] Step S3 is mainly used to measure the raw output value of the sensor when a material is placed in the tray, and to calculate the calibration value. After the initial value measurement is completed, a material is placed in the tray and heated at a constant temperature for a period of time (e.g., 20 minutes). Then, n output data points from each sensor are collected at a collection period T (e.g., 1 second) to obtain n raw output values ​​x'. i , using x' i Subtract the initial value x 0i Obtain the calibration value x after eliminating the influence of the initial value. i =x' i -x 0i .

[0095] Step S4 is mainly used to repeat step S3 after changing the material until the calibration value of the sensor output data is obtained when the three materials are placed in the tray respectively.

[0096] Step S5 is mainly used to measure the calibration values ​​of the sensor output data at different temperatures. In this embodiment, by increasing the heating stage temperature in steps, steps S2 to S5 are repeated at each temperature point to obtain the calibration values ​​of the sensor output data when different materials are placed at each temperature point. The calibration values ​​of the sensor output data form a dataset S, which is an N-row, 6-column matrix S0. N×6 As shown in the matrix expression above. The size of N is equal to the product of the number of material types (3), the number of temperature points (m), and the number of data points collected at each temperature point (n), 3mn. Sometimes, in order to improve the experimental accuracy, the experiment may need to be repeated, and the number of N may be doubled.

[0097] PVC undergoes significant pyrolysis (thermal decomposition) and releases a large amount of volatile gases when the temperature reaches above 260℃, while releasing almost no volatile gases at temperatures below 130℃. Therefore, the heating temperature range for PVC was set to 130–260℃. Similarly, the heating temperature range for epoxy resin and ABS was set to 150–310℃. During the experiment, 27 temperature points were uniformly selected within the heating temperature ranges for PVC, ABS, and epoxy resin, respectively, and these three materials were heated. To eliminate gross errors, two independent experiments were conducted at each heating temperature point; if the difference between the two experiments was too large, the experiment was repeated. After the experiment, a total of 162 experiments were conducted, of which 66 groups simulated the non-overheated state of the electrical equipment, 32 groups simulated the overheated state of the ABS electrical equipment, 32 groups simulated the overheated state of the epoxy resin electrical equipment, and 32 groups simulated the overheated state of the PVC electrical equipment. When 71 data points were collected at each temperature point, the number of rows in the resulting dataset S was N = 162 × 71 = 11502.

[0098] As an optional embodiment, the method further includes: first normalizing each data point in S0, and then dividing the normalized S0 into S1 and S2; the normalization formula is:

[0099]

[0100] In the formula, x i - For any data x in the i-th column of S0 i The normalized value, x i-min x i-max Let i and y be the minimum and maximum values ​​of the i-th column of S0, respectively, where i = 1, 2, ..., 6.

[0101] This embodiment provides a data preprocessing technical solution. Before dividing S0 into S1 and S2, this embodiment first normalizes S0 according to the formula above, converting the data in S0 into [0,1] data. According to the formula above, when x i =x i-min When, the normalized value x i - =0; when x i =x i-max When, the normalized value x i - =1; other x i normalized value x i - Satisfy 0 <x i - <1.

[0102] As an optional embodiment, the method further includes: performing dimensionality reduction processing on the training dataset S1 and the test dataset S2, and training and testing the detection model based on the dimensionality-reduced training dataset and test dataset.

[0103] This embodiment presents a technical solution to improve the prediction accuracy of a detection model. Because the datasets S1 and S2 obtained in this embodiment have high dimensionality, large data volume, and indistinct data features, directly using these data as input to the detection model results in a large amount of data for training the model and low prediction accuracy. Therefore, this embodiment first performs dimensionality reduction processing on the data in datasets S1 and S2, and then uses the dimensionality-reduced training and testing datasets to train and test the detection model. Many existing data dimensionality reduction algorithms exist; this embodiment does not limit the specific dimensionality reduction algorithm, but a specific dimensionality reduction algorithm will be given in later embodiments.

[0104] As an optional embodiment, the method for dimensionality reduction of the training dataset S1 includes:

[0105] Running the PCA algorithm on S1 yields a linear transformation matrix W1 with 1 rows and 6 columns, where w1 = 1;

[0106] Running the LDA algorithm on S1 yields a linear transformation matrix W2 with 6 columns and 2 rows, where 1 ≤ w2 ≤ 3;

[0107] Combine W1 and W2 into a matrix W with 6 rows and 6 columns, consisting of w1+w2 rows.

[0108] Dimensionality reduction of S1 is performed using the following formula:

[0109] S 1- =W×(S1) T

[0110] In the formula, S1- This is the matrix after S1 has been reduced in dimension.

[0111] This embodiment provides a technical solution for dimensionality reduction of training dataset S1. This technical solution is also suitable for dimensionality reduction of S2.

[0112] This embodiment proposes a data dimensionality reduction scheme combining PCA and LDA algorithms to extract features strongly correlated with temperature and material, respectively. Since the four data categories in S1 are jointly determined by temperature and material, extracting features strongly correlated with temperature and material from S1 can effectively improve the prediction accuracy of the detection model. Practice shows that temperature has the most significant impact on sensor output and is also the main reason for differences between different data. Therefore, features strongly correlated with temperature should reflect the differences between different data parameters to the greatest extent possible. The PCA algorithm can extract a set of principal components with the largest variance from S1. These principal components can reflect the differences in data parameters within S1 to the greatest extent possible. Since both principal components and temperature-correlated features can reflect the differences in data parameters within S1 well, it is reasonable to use the principal components extracted by the PCA algorithm as temperature-correlated features. However, the PCA algorithm aims to maximize the variance of the extracted principal components, resulting in non-selective extraction. This makes it very difficult to use the PCA algorithm to selectively extract features strongly correlated with material. The LDA algorithm utilizes class information to project high-dimensional data samples onto the optimal discriminant vector space, ensuring that the projected features satisfy the conditions of minimum intra-class variation and maximum inter-class variation. Therefore, LDA can be used to extract features strongly correlated with materials. However, LDA can only handle data with a limited number of classes. Since temperature is a continuous quantity, the extracted parameters strongly correlated with temperature should also be continuous. Dividing temperature into a finite number of classes would minimize intra-class variation in the extracted results, reducing the correlation between the extracted results and temperature. Therefore, this embodiment combines PCA and LDA algorithms to achieve data dimensionality reduction. The specific technical solution is described below.

[0113] First, the PCA algorithm is run on S1 to obtain a linear transformation matrix W1 with rows w1 and columns 6. The PCA algorithm is a well-established dimensionality reduction algorithm; the covariance matrix S1(S1) of S1 is calculated first. T Then for S1(S1) T Perform eigenvalue decomposition to obtain S1(S1). T The eigenvectors and eigenvalues ​​are obtained, and the eigenvectors corresponding to the largest w1 eigenvalues ​​are selected to obtain the linear transformation matrix W1. In this embodiment, W1 is used to extract features that are strongly correlated with the variable temperature, therefore w1 = 1.

[0114] Secondly, the LDA algorithm is run on S1 to obtain a linear transformation matrix W2 with 2 rows and 6 columns. The LDA algorithm is also a mature dimensionality reduction algorithm. It first transforms the original S1, which contains 4 categories, into 3 matrices S1 corresponding to 3 categories (epoxy resin overheating, PVC overheating, and ABS overheating). A S B S C Then calculate the within-class scatter matrix S of S1. w and the inter-class scatter matrix S b Then, the dimensionality reduction matrix S is calculated. e =S w ×(S b ) T Then for S e Eigenvalue decomposition is performed to obtain eigenvalues ​​and eigenvectors. The eigenvectors corresponding to the two largest eigenvalues ​​(w2) are selected to obtain the linear transformation matrix W2. In this embodiment, W2 is used to extract features strongly correlated with the material. Since the material includes three categories, W2 satisfies 1 ≤ w2 ≤ 3.

[0115] Next, combine W1 and W2 into a matrix W with 6 rows and 6 columns, i.e.:

[0116]

[0117] Finally, the dimensionality of S1 is reduced by multiplying matrix W with the transpose of S1.

[0118] It is worth noting that determining the size of w2 is crucial. If it is too large, the dimensionality reduction effect will be insignificant, failing to meet the requirement of improving the prediction accuracy of the detection model; if it is too small, effective classification may not be achieved. In this embodiment, 1≤w2≤3. Since the goal is dimensionality reduction, w2 can only take the values ​​1 or 2. When w2 = 1, effective classification may not be achieved. Therefore, when running the LDA algorithm, the eigenvectors corresponding to the two largest eigenvalues ​​are selected.

[0119] As an optional embodiment, the detection model is a second-order BP neural network consisting of a material identification network and an overheating diagnosis network; the material identification network is used to determine the category of the heated material, and the overheating diagnosis network is used to output the overheating detection result.

[0120] This embodiment presents a network structure for the detection model. Since this embodiment aims to achieve both overheat detection and overheat material category identification, the detection model is designed as a second-order backpropagation (BP) neural network consisting of a material identification network and an overheat diagnosis network. The material identification network has a total of three layers: one input layer, one output layer, and one hidden layer, with 3, 3, and 10 neurons in each layer. The overheat diagnosis network also consists of one input layer, one output layer, and one hidden layer, with 4, 4, and 10 neurons in each layer.

[0121] To demonstrate the feasibility of the method described in this invention, the accuracy of using this method for diagnosing overheating in electrical equipment is presented below. The diagnostic accuracy rates of the method for four conditions—non-overheating, ABS overheating, PVC overheating, and epoxy resin overheating—are 93.64%, 99.57%, 89.78%, and 93.91%, respectively, with a 100% accuracy rate for identifying overheated materials. This demonstrates that the method described in this invention can effectively identify overheating faults in electrical equipment.

[0122] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting overheating in electrical equipment based on a sensor array, characterized in that, Includes the following steps: Acquire output data from sensors in a sensor array, wherein the sensors are used to measure the concentrations of different volatile gases generated in the working environment of electrical equipment; The sensor's output data is input into a detection model based on a BP neural network to obtain the electrical equipment overheating detection results: no overheating, epoxy resin overheating, PVC overheating, ABS overheating; An alarm signal is issued when the detection result indicates that the epoxy resin, PVC, or ABS is overheated, and the duration exceeds the set threshold. Wherein: the sensor includes: The first sensor is used to detect hydroxyl-containing compounds in volatile gases that have properties similar to alcohol; The second sensor is used to detect flammable components in volatile gases that are similar in properties to methane and hydrogen. The third sensor is used to detect volatile organic compounds in gaseous volatiles that are similar in properties to benzene and phenol; The fourth sensor is used to detect methyl-containing components in volatile gases that have properties similar to methane; The fifth sensor is used to detect methyl-containing components in volatile gases that have properties similar to ethane; The sixth sensor is used to detect small molecule components in volatile gases that are similar in properties to natural gas. The method for training the detection model includes: Data set S was obtained through simulation experiments. The data in S consisted of sensor output data at different temperatures when different types of materials were used to simulate the different volatile gases. The types of materials included epoxy resin, PVC, and ABS. Each material has an overheating temperature threshold. By comparing the temperature corresponding to each data point in S with the overheating temperature threshold of the corresponding material, a detection result label is obtained for each data point. The detection result label includes no overheating, epoxy resin overheating, PVC overheating, and ABS overheating. Each data point in S is labeled based on the obtained detection result labels to obtain a dataset S0 including the detection result labels; Half of the sample data is extracted from S0 to obtain the training dataset S1, and the other half of the sample data forms the test dataset S2; The detection model is trained using the training dataset S1, and the trained detection model is tested using the test dataset S2. The detection model is a second-order BP neural network consisting of a material identification network and an overheating diagnosis network; the material identification network is used to determine the category of the heated material, and the overheating diagnosis network is used to output the overheating detection result.

2. The method for overheat detection of electrical equipment based on a sensor array according to claim 1, characterized in that, The first to sixth sensors are TGS822, TGS813, TGS2602, TGS2620, TGS2610 and TGS2611, respectively.

3. The method for overheat detection of electrical equipment based on a sensor array according to claim 2, characterized in that, The sensor is a metal oxide gas sensor composed of gas-sensitive elements. The resistance of the gas-sensitive elements changes with the concentration of the detected gas; the higher the concentration of the detected gas, the lower the resistance of the gas-sensitive elements. The output voltage of the sensor is: In the formula, V is the output voltage of the sensor, and K S R is the signal amplification factor. S R0 is the resistance of the gas-sensitive element, C is the reference resistance, and C is an inherent constant.

4. The method for overheat detection of electrical equipment based on a sensor array according to claim 1, characterized in that, The method for obtaining dataset S through simulation experiments includes: S1. A simulation experimental device that connects a sensor array, a data acquisition module, a host computer, and a heating stage, wherein the sensor array and the heating stage are housed in the same cavity; S2. Set the temperature of the heating platform. After the temperature stabilizes, read the output data of each sensor in the sensor array to obtain the initial value of the sensor. The initial value of the i-th sensor is represented as x. 0i Where i = 1, 2, ..., 6; S3. Place a type of material on the tray of the heating table, heat it at a constant temperature for a period of time, and read n output data from each sensor according to the acquisition period T. The output data of the i-th sensor is x'. i Calculate x i =x' i -x 0i The categories of materials include epoxy resin, PVC, and ABS. S4. Release the remaining gas in the cavity, then proceed to step S3 and replace the material on the tray with another material; continue until all three materials have been replaced. S5. Increase the heating stage temperature by a step size, and repeat steps S2 to S5 until the heating stage temperature reaches the set upper temperature limit, resulting in a dataset S represented in the following matrix form: In the formula, N = 3 × m × n, where m is the number of temperature points set on the heating platform.

5. The method for overheat detection of electrical equipment based on a sensor array according to claim 4, characterized in that, The method further includes: first normalizing each data point in S0, and then dividing the normalized S0 into S1 and S2; the normalization formula is: In the formula, x i - For any data x in the i-th column of S0 i The normalized value, x i-min x i-max Let i and y be the minimum and maximum values ​​of the i-th column of S0, respectively, where i = 1, 2, ..., 6.

6. The method for overheat detection of electrical equipment based on a sensor array according to claim 4, characterized in that, The method further includes: performing dimensionality reduction processing on the training dataset S1 and the test dataset S2, and training and testing the detection model based on the dimensionality-reduced training dataset and test dataset.

7. The method for overheat detection of electrical equipment based on a sensor array according to claim 6, characterized in that, Methods for dimensionality reduction of the training dataset S1 include: Running the PCA algorithm on S1 yields a linear transformation matrix W1 with 1 rows and 6 columns, where w1 = 1; Running the LDA algorithm on S1 yields a linear transformation matrix W2 with 6 columns and 2 rows, where 1 ≤ w2 ≤ 3; Combine W1 and W2 into a matrix W with 6 rows and 6 columns, consisting of w1+w2 rows. Dimensionality reduction of S1 is performed using the following formula: S 1- =W×(S1) T In the formula, S 1- This is the matrix after S1 has been reduced to its original dimensions.

Images