GIS equipment moss detection method based on temperature and humidity prediction and image segmentation

Abstract

The invention discloses a GIS equipment moss detection method based on temperature and humidity prediction and image segmentation. The method comprises the following steps: acquiring environmental data of a transformer substation where gas insulated switch GIS equipment is located and preprocessing the environmental data; correlation analysis is carried out on the preprocessed environment data, and characteristic parameters of temperature and humidity prediction are determined; establishing a temperature and humidity prediction model by adopting a neural network architecture, and inputting the determined characteristic parameters into the temperature and humidity prediction model to obtain temperature and humidity prediction data; constructing a temperature and humidity weighted scoring model, performing time sequence weighted processing on the temperature and humidity prediction data, and screening a target transformer substation; and acquiring a GIS equipment surface image of the target substation, performing curve segmentation on the GIS equipment surface image, and detecting a moss distribution position on the surface of the GIS equipment. The problems that a method for detecting moss on the surface of GIS equipment mainly depends on manual visual inspection, efficiency is low, real-time detection cannot be achieved, temperature and humidity environment parameters are neglected, and moss growth prediction capacity is lacked are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent monitoring and maintenance of power equipment, and discloses a moss detection method for GIS equipment based on temperature and humidity prediction and image segmentation. Background Art

[0002] In the power system, as a core device to ensure the safe operation of the power grid, the gas-insulated switchgear (GIS) is long-term exposed to the complex and changeable outdoor environment. The surface of the GIS is susceptible to environmental factors and moss is likely to grow. The growth of moss will not only reduce the insulation performance of the equipment, accelerate the corrosion of metal components, but also may cause partial discharge or short-circuit faults, seriously threatening the stability and safety of the power system.

[0003] The traditional moss detection on the surface of GIS equipment mainly relies on manual visual inspection, and the operation and maintenance personnel need to check the surface status of the equipment station by station. However, manual inspection has problems of low efficiency and limited coverage, especially it is difficult to achieve comprehensive detection in a large-scale substation network. In addition, the detection results are easily affected by subjective factors such as differences in personnel experience and visual fatigue, resulting in a high risk of missed detection or misjudgment, and it is impossible to track the growth trend of moss in real time, making it difficult to timely warn of potential safety hazards.

[0004] In addition, when dealing with images in complex environments, the existing detection methods often do not fully consider the dynamic environmental conditions for moss growth, and cannot effectively adapt to different lighting conditions, weather conditions and complex backgrounds on the equipment surface, making it difficult to achieve dynamic assessment of the moss growth potential, resulting in insufficient reliability and prediction ability of the detection results. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems in the existing moss detection method for the surface of GIS equipment, which mainly relies on manual visual inspection, has low efficiency and cannot be monitored in real time, ignores environmental parameters such as temperature and humidity, has a high misjudgment rate in complex scenarios, and lacks the ability to predict the growth of moss.

[0006] To solve the above technical problems, the technical solution provided by the present invention is a moss detection method for GIS equipment based on temperature and humidity prediction and image segmentation, including the following steps: Step 1: Collect the environmental data of the substation where the gas-insulated switch (GIS) equipment is located and preprocess the environmental data; the environmental data includes temperature and humidity, wind speed, wind chill index, heat index, solar radiation, ultraviolet index and 1-minute precipitation, and the temperature and humidity include temperature and humidity; The preprocessing includes normalizing the environmental data; Step 2: Conduct a correlation analysis on the environmental data preprocessed in Step 1 to determine the characteristic parameters for temperature and humidity prediction; The characteristic parameters for determining the temperature and humidity prediction are analyzed for correlation using the Pearson correlation coefficient; The datasets of temperature and humidity are , that is ; The datasets of wind speed, wind chill index, heat index, solar radiation, ultraviolet index, and 1-minute precipitation are , that is ; The calculation formula for the Pearson correlation coefficient r value is: ; In the formula, is the numerical mean, is the numerical mean, and n is the length of the time series; The influencing factors with an absolute value of the Pearson correlation coefficient r lower than 0.2 are excluded The characteristic parameters for temperature and humidity prediction include temperature and humidity, wind speed, wind chill index, heat index, solar radiation, ultraviolet index, and 1-minute precipitation.

[0007] Step 3: Establish a temperature and humidity prediction model using a neural network architecture, and input the characteristic parameters determined in Step 2 into the temperature and humidity prediction model to obtain temperature and humidity prediction data; The neural network architecture is a CGC-BILSTM multi-task deep neural network architecture. The input is the characteristic parameters determined in Step 2, and the output is the predicted temperature and humidity data, specifically including: A bidirectional long short-term memory network BiLSTM layer, which is used to extract the temporal relationship of the input features; A shared network, which consists of a convolutional neural network and a splicing layer, and is used to fuse the features of multiple expert networks; A gating network, which generates a weight vector through the Sigmoid function and dynamically weights the output of the expert network; A tower network, which outputs the temperature and humidity prediction results through the ReLU activation function and linear transformation; The input data passes through the BiLSTM layer and outputs temporal features; The calculation formula for the BiLSTM layer is: ; In the formula, is the input sequence, is the output state, is the hidden state of the forward LSTM at time step t-1, is the hidden state of the backward LSTM at time step t-1; Connect the expert network and the shared network, and the output state of the BiLSTM layer As input features, they are concatenated in the parameter dimension and then enter the shared network; The formula of the shared network is: ; In the formula, is the output of the th expert network, is the shared feature, CNN( ) is the convolutional neural network, Concat( ) is to concatenate multiple vectors in the dimension; Weight distribution is performed through the gating network; The output of the gating network is the weight vector, and the calculation formula of the weight vector is: ; In the formula, is the Sigmoid function, and are learning parameters; The weighted feature is: ; The output layer is obtained through the tower network, and the weighted feature is further processed to output the prediction result; The calculation formula of the prediction result is: ; In the formula, ReLU( ) is the activation function, Linear( ) is to perform a linear transformation on the feature processed by the ReLU activation function; The calculation formula of the loss function is: ; In the formula, is the single-task loss, is the task weight coefficient, is the mean square error, is the true value of the i-th task.

[0008] Step 4: Construct a temperature and humidity weighted scoring model, perform time series weighted processing on the temperature and humidity prediction data in Step 3, and screen out the target substations according to the preset threshold of moss growth conditions; Step 4.1: According to the output results of the temperature and humidity prediction model in Step 3, collect the temperature and humidity data of the substation for the next 7 days; Step 4.2: Use the temperature and humidity weighted comprehensive screening method to construct a temperature and humidity weighted scoring model and calculate the final comprehensive score ; Step 4.2.1: Set the grading rules for temperature T and humidity H, and calculate the temperature score of the substation and the humidity score ; The temperature score and the humidity score The calculation formulas are as follows: ; .

[0009] Step 4.2.2: Calculate the time weight according to the consecutive days P that meet the moss growth conditions ; The time weight The calculation formula is: .

[0010] Step 4.2.3: Calculate the daily comprehensive score , and calculate the final comprehensive score according to the time weight of the consecutive days P ; The final comprehensive score The calculation formula is: ; ; In the formula, is the temperature weight, is the humidity weight.

[0011] Step 4.3: Set the threshold of moss growth conditions according to the final comprehensive score S, and determine the moss growth risk range of the substation; The moss growth condition threshold specifically includes: .

[0012] Step 4.4: Obtain the filtered target substation.

[0013] Step 5: Obtain the surface image of the GIS equipment of the target substation, perform curve segmentation on the surface image of the GIS equipment, and detect the moss distribution position on the surface of the GIS equipment.

[0014] Step 5.1: Collect the surface image data of the GIS equipment of the target substation obtained in Step 4 and convert the data from the RGB color space to the HSI color space; Step 5.2: For the HSI color space obtained in Step 5.1, construct a cyclic energy curve CEC on the H component of the HSI color space and calculate the pixel value energy of the image; The formula for calculating the pixel value energy of the cyclic energy curve CEC is as follows: ; In the formula, is the pixel value, is the neighborhood system of the pixel; Step 5.3: Use the Tsallis entropy to find the segmentation threshold K of the CEC. The formula for the Tsallis entropy is: ; In the formula, is the parameter of the Tsallis entropy, is the probability distribution of the pixel; Step 5.4: Segment the linearized CEC according to the segmentation threshold K in Step 5.3; Step 5.5: Detect the moss position on the surface of the GIS device according to the segmentation result in Step 5.4.

[0015] Compared with the prior art, the beneficial effects of the present invention include: (1) By integrating multi-source environmental parameters such as temperature and humidity, wind speed, solar radiation, ultraviolet index, and precipitation, the present invention dynamically captures the non-linear synergistic effect of multiple factors. Combining with the CGC-BILSTM multi-task neural network architecture, the bidirectional long short-term memory network BiLSTM is used to extract temporal features, and the gating network is used to dynamically allocate the weights of the expert network to achieve high-precision temperature and humidity prediction. On this basis, a temperature and humidity weighted scoring model is constructed, and the temperature and humidity scores are quantified according to the piecewise function, combined with the time series weights of consecutive days, to scientifically screen out the substations with high moss incidence, breaking through the limitations of the traditional method in terms of single use of environmental parameters, rigid screening rules, and waste of computing resources.

[0016] (2) By constructing a temperature and humidity weighted scoring model, the present invention breaks through the defects of the traditional binary threshold or fixed weight combination, and dynamically quantifies the driving effect of environmental parameters on moss growth. The temperature score is assigned points according to the biological optimal interval in segments, and the humidity score distinguishes the short-term fluctuation and the long-term high humidity influence, comprehensively and systematically evaluating the temperature and humidity data of the substation. Combining with the dynamic weights of consecutive days of compliance, the screening accuracy is improved, and the high-risk substations are accurately locked, providing a high-value target set for subsequent image processing.

[0017] (3)The processing of the surface image of the GIS device in the present invention directly correlates with the differential features of moss and complex backgrounds. By converting the RGB image to the HSI color space and utilizing the characteristic that the hue H component is insensitive to illumination, a cyclic energy curve (CEC) is constructed, the pixel neighborhood difference is defined, and the moss edge features are enhanced. Further, Tsallis entropy threshold segmentation is adopted to break through the dependence on linear assumptions of traditional threshold methods, predict and detect the growth of moss on the surface of the GIS device, effectively avoid the influence of moss attachment on the device, and prevent potential safety hazards caused thereby. Description of the Drawings

[0018] The present invention will be further described below in conjunction with the drawings and embodiments.

[0019] Figure 1 It is a schematic diagram of the flow of the moss detection method for GIS devices based on temperature and humidity prediction and image segmentation according to an embodiment of the present invention.

[0020] Figure 2 It is a flowchart of the Pearson correlation analysis according to an embodiment of the present invention.

[0021] Figure 3 It is a flowchart of the temperature and humidity prediction of the CGC - BILSTM multi - task deep neural network architecture according to an embodiment of the present invention.

[0022] Figure 4 It is a flowchart of the temperature and humidity weighted scoring model according to an embodiment of the present invention.

[0023] Figure 5 It is a flowchart of the color image segmentation based on the Tsallis entropy - based linearized cyclic energy curve according to an embodiment of the present invention. Detailed Embodiments

[0024] As Figure 1 and Figure 2 and Figure 3 and Figure 4 and Figure 5 shown, the moss detection method for GIS devices based on temperature and humidity prediction and image segmentation includes the following steps: Step 1: Collect the environmental data of the substation where the gas - insulated switch (GIS) device is located and pre - process the environmental data; the environmental data includes temperature and humidity, wind speed, wind chill index, heat index, solar radiation, ultraviolet index, and 1 - minute precipitation amount, and the temperature and humidity include temperature and humidity; The pre - processing includes normalizing the environmental data; Step 2: Perform correlation analysis on the pre - processed environmental data in Step 1 using the Pearson correlation coefficient to determine the characteristic parameters for temperature and humidity prediction; Due to the large number of environmental factors affecting temperature and humidity near the GIS equipment in the substation, taking all of them into account in the analysis of the temperature and humidity change law and prediction calculation will bring a large amount of unnecessary workload. In fact, the degrees of influence of relevant environmental factors on temperature and humidity are different. Therefore, before predicting temperature and humidity, a correlation analysis is carried out on the data input into the prediction model, and the influence factors with weak correlation are eliminated to reduce the model calculation complexity. The Pearson correlation coefficient can accurately reflect the correlation between variables. In this embodiment, this coefficient is used to judge the relationship between temperature and humidity and their influencing factors.

[0025] The data sets of temperature and humidity are , that is ; The data sets of wind speed, wind chill index, heat index, solar radiation, ultraviolet index and 1-minute precipitation are , that is ; The calculation formula of the Pearson correlation coefficient r value is: ; In the formula, is the numerical mean value, is the numerical mean value, and n is the length of the time series; r is used to describe and the degree of linear correlation between , the greater and the stronger the degree of linear correlation. If , it reflects that and are in a positive correlation relationship; if , it means that and are not correlated; if , it reflects that and are in a negative correlation relationship.

[0026] The Pearson correlation coefficients are calculated for the collected data of wind speed, wind chill index, heat index, solar radiation, ultraviolet index and 1-minute precipitation respectively with temperature and humidity to screen and determine the relevant influencing factors of temperature and humidity.

[0027] In the calculation process, the influencing factors with the absolute value of the Pearson correlation coefficient lower than 0.2 are eliminated. The characteristic descriptions of each relevant data are shown in Table 1;

[0028] The influencing factors with the absolute value of the Pearson correlation coefficient r lower than 0.2 are eliminated The characteristic parameters for temperature and humidity prediction include temperature and humidity, wind speed, wind chill index, heat index, solar radiation, ultraviolet index, and 1-minute precipitation.

[0029] Step 3: Establish a temperature and humidity prediction model using a neural network architecture. Input the characteristic parameters determined in Step 2 into the temperature and humidity prediction model to obtain temperature and humidity prediction data. The neural network architecture is a CGC-BILSTM multi-task deep neural network architecture. Input the characteristic parameters determined in Step 2, and the output is the predicted temperature and humidity data, including: Bidirectional Long Short-Term Memory Network (BiLSTM) layer, which is used to extract the temporal relationship of the input features. Shared network, which consists of a convolutional neural network and a concatenation layer, and is used to fuse the features of multiple expert networks. Gating network, which generates a weight vector through the Sigmoid function to dynamically weight the output of the expert network. Tower network, which outputs the temperature and humidity prediction results through the ReLU activation function and linear transformation. The input data passes through the BiLSTM layer, and the output is the temporal features. The calculation formula of the BiLSTM layer is: ; In the formula, is the input sequence, is the output state, is the hidden state of the forward LSTM at time step t-1, is the hidden state of the backward LSTM at time step t-1; Connect the expert network and the shared network. The output state of the BiLSTM layer is used as the input feature, and after concatenation in the parameter dimension, it enters the shared network. The formula of the shared network is: ; In the formula, is the output of the th expert network, is the shared feature, CNN( ) is the convolutional neural network, and Concat( ) is to concatenate multiple vectors in the dimension; Perform weight allocation through the gating network. The output of the gating network is the weight vector, and the calculation formula of the weight vector is: ; In the formula, is the Sigmoid function, and is a learning parameter; The weighted feature is: ; The output layer is obtained through the tower network, and the weighted feature is further processed to output the prediction result; The calculation formula of the prediction result is: ; In the formula, ReLU( ) is the activation function, and Linear( ) performs a linear transformation on the feature processed by the ReLU activation function; The calculation formula of the loss function is: ; In the formula, is the single-task loss, is the task weight coefficient, is the mean square error, is the true value of the i-th task.

[0030] Finally, the predicted temperature and humidity data are output, and the temperature and humidity prediction model is used to predict the temperature and humidity data for the next 7 days, which is used for the temperature and humidity weighted scoring model to screen the target substation.

[0031] To verify the superiority of the CGC-BiLSTM multi-task deep neural network architecture proposed in this embodiment for establishing a temperature and humidity prediction model, the neural network architecture of the existing technology is selected as the comparison model. In the temperature prediction tasks of different substations, different substations are selected as Task 1, Task 2, and Task 3 respectively. The mean absolute error MAE, the coefficient of variation root mean square error CV-RMSE, and the coefficient of determination R² are used as evaluation indicators. The calculation of the evaluation indicators is shown in Table 2:

[0032] Based on the experimental data verification shown in Table 2, the CGC-BiLSTM proposed by the present invention shows significant technical advantages in the temperature and humidity prediction of substations.

[0033] Step 4: Construct a temperature and humidity weighted scoring model, perform time series weighting processing on the temperature and humidity prediction data in Step 3, and screen out the target substation according to the preset threshold of the moss growth conditions; The growth characteristics of moss are as follows: it has a high growth tendency in a cool, humid environment with an acidic pH value below 5. The growth cycle of moss is relatively short, and a relatively dense green algae layer can be formed within 3 to 15 days. The specific growth rate depends on environmental conditions. When the relative air humidity exceeds 80% and the environmental temperature is between 22°C and 25°C, the growth rate of bryophytes significantly increases, and it is extremely easy to form a relatively large community. Under the condition that the temperature reaches 35°C, the growth of green algae will reach its peak within 4 to 5 days. Further, when the relative air humidity continuously remains higher than 80% and the environmental temperature is maintained between 25°C and 30°C, after a 7-day incubation period, the number of green algae will increase rapidly and reach its peak on the 8th to 9th day. In addition, based on single-factor experiments on the growth characteristics of moss, its most suitable growth temperature range is 25°C to 35°C, and the relative humidity is between 90% and 100%.

[0034] Regarding the distribution of moss, taking the suspension of the tension string of a glass insulator as an example, it is observed that the distribution density of moss on the upper surface of the insulator is significantly lower than that on the lower surface. Specifically, the distribution of moss is relatively concentrated in the areas along the outer edge of the upper side rib and the inner edge of the lower side rib on the lower surface of the insulator.

[0035] When determining that the temperature weight is greater than the humidity weight, the growth characteristics of moss, the actual environmental impact, and data sensitivity analysis are comprehensively considered. Temperature is more restrictive to the growth of moss. Moss can only grow well between 22°C and 35°C, and its growth rate is the fastest at 25°C to 30°C. The suitable range of humidity is wider. As long as the relative humidity is above 80%, moss can grow. Temperature has a greater impact on the growth cycle of moss, directly affecting its growth rate and reproduction cycle, while humidity mainly plays an auxiliary role. In the actual environment, the impact of temperature change on the growth of moss is more direct and significant. Unsuitable temperature will directly inhibit the growth of moss, while the impact of humidity change is relatively lagging. Temperature data is more sensitive and volatile to the detection of moss growth, and can more timely and accurately reflect the possibility of moss growth. Therefore, in this study, the temperature weight is set to 0.6 and the humidity weight is set to 0.4 to more scientifically evaluate the risk of moss growth on the surface of substation GIS equipment.

[0036] Step 4.1: According to the output results of the temperature and humidity prediction model in Step 3, collect the temperature and humidity data of the substation for the next 7 days; Step 4.2: Construct a temperature and humidity weighted scoring model using the temperature and humidity weighted comprehensive screening method to calculate the final comprehensive score ; Step 4.2.1: Set the grading scoring rules for temperature T and humidity H, and calculate the temperature score of the substation and humidity score ; Temperature score And humidity score The calculation formulas are as follows: ; .

[0037] Step 4.2.2: Calculate the time weight according to the consecutive days P that meet the moss growth conditions ; Time weight The calculation formula is: .

[0038] Step 4.2.3: Calculate the daily comprehensive score , and calculate the final comprehensive score according to the time weight of the consecutive days P ; Final comprehensive score The calculation formula is: ; ; In the formula, is the temperature weight with a value of 0.6, is the humidity weight with a value of 0.4.

[0039] Step 4.3: Set the moss growth condition threshold according to the final comprehensive score S, and determine the moss growth risk range of the substation; The moss growth condition threshold specifically includes: .

[0040] Step 4.4: Obtain the filtered target substation.

[0041] The temperature and humidity weighted scoring model not only considers whether the temperature and humidity are within the range suitable for moss growth, but also more scientifically evaluates the overall suitability of the substation through the weighted comprehensive scoring method, effectively improving the accuracy and scientificity of the screening. At the same time, the introduction of time series weighting makes the screening method more comprehensive and can better reflect the influence of temperature and humidity conditions for consecutive days on moss growth.

[0042] In this way, initially screen out which substations in this area have an environment suitable for moss growth, discard the substations that are not suitable for moss growth, and reduce the number of substations in this area that need to perform color image segmentation to find the moss position, so as to reduce the large amount of image segmentation calculation.

[0043] Step 5: Obtain the surface image of the GIS equipment of the target substation, perform curve segmentation on the surface image of the GIS equipment, and detect the moss distribution position on the surface of the GIS equipment.

[0044] Step 5.1: Collect the surface image data of the GIS equipment of the target substation obtained in Step 4 and convert the data from the RGB color space to the HSI color space; Step 5.2: For the HSI color space obtained in Step 1, construct a cyclic energy curve CEC on the H component of the HSI color space and calculate the pixel value energy of the image; The formula for calculating the pixel value energy of the cyclic energy curve CEC is: ; In the formula, is the pixel value, is the neighborhood system of the pixel; Step 5.3: Use the Tsallis entropy to find the segmentation threshold K of the CEC. The formula for the Tsallis entropy is: ; In the formula, is the parameter of the Tsallis entropy, is the probability distribution of the pixel; Step 5.4: Segment the linearized CEC according to the segmentation threshold K in Step 5.3; Step 5.5: Detect the moss position on the surface of the GIS equipment according to the segmentation result in Step 5.4.

[0045] Through the above color image segmentation method of the linearized cyclic energy curve based on the Tsallis entropy, the growth situation and specific position of the moss on the surface of the GIS equipment of the screened substation in this area can be detected efficiently and accurately, effectively preventing potential safety hazards and improving the efficiency of the substation patrol work.

Claims

1. A method for detecting moss in GIS equipment based on temperature and humidity prediction and image segmentation, characterized in that, It includes the following steps: Step 1: Collect the environmental data of the substation where the gas-insulated switch GIS equipment is located and preprocess the environmental data; Step 2: Conduct a correlation analysis on the environmental data preprocessed in Step 1 to determine the characteristic parameters for temperature and humidity prediction; Step 3: Establish a temperature and humidity prediction model using a neural network architecture, input the characteristic parameters determined in Step 2 into the temperature and humidity prediction model, and obtain the temperature and humidity prediction data; Step 4: Construct a temperature and humidity weighted scoring model, perform time series weighting on the temperature and humidity prediction data in Step 3, and screen out the target substations according to the preset threshold of moss growth conditions; Step 5: Obtain the surface image of the GIS equipment of the target substation, perform curve segmentation on the surface image of the GIS equipment, and detect the distribution position of moss on the surface of the GIS equipment.

2. The moss detection method for GIS equipment based on temperature and humidity prediction and image segmentation according to claim 1, characterized in that, In Step 1, the environmental data includes temperature and humidity, wind speed, wind chill index, heat index, solar radiation, ultraviolet index, and 1-minute precipitation, and the temperature and humidity include temperature and humidity.

3. The moss detection method for GIS equipment based on temperature and humidity prediction and image segmentation according to claim 2, characterized in that, In Step 2, to determine the characteristic parameters for temperature and humidity prediction, a Pearson correlation coefficient is used for correlation analysis; The data set of temperature and humidity is , namely ; The data sets of wind speed, wind chill index, heat index, solar radiation, ultraviolet index, and 1-minute precipitation are , namely ; The calculation formula for the Pearson correlation coefficient r value is: ； In the formula, is the numerical mean, is the numerical mean, and n is the length of the time series; The characteristic parameters for temperature and humidity prediction include temperature and humidity, wind speed, wind chill index, heat index, solar radiation, ultraviolet index, and 1-minute precipitation.

4. The method for detecting moss on GIS equipment based on temperature and humidity prediction and image segmentation according to claim 3, characterized in that, In Step 3, the neural network architecture is the CGC-BILSTM multi-task deep neural network architecture. Input the characteristic parameters determined in Step 2, and the output is the predicted temperature and humidity data; The input data passes through the BiLSTM layer, and the output is the time series characteristics; The calculation formula for the BiLSTM layer is: ； Wherein, is the input sequence, is the output state, is the hidden state of the forward LSTM at time step t - 1, is the hidden state of the backward LSTM at time step t - 1; Connect the expert network and the shared network, and the output state of the BiLSTM layer As input features, they are concatenated in the parameter dimension and then enter the shared network; The formula for the shared network is: ； In the formula, is the output of the th expert network, is the shared feature, CNN( ) is a convolutional neural network, and Concat( ) is to concatenate multiple vectors in dimension; Weight distribution is performed through the gating network; The output of the gating network is the weight vector, and the calculation formula for the weight vector is: ； wherein, is the Sigmoid function, and are learning parameters; The weighted feature is: ； The output layer is obtained through the tower network, and the weighted feature is further processed and the prediction result is output; The calculation formula for the prediction result is: ； where ReLU( ) is the activation function, and Linear( ) performs a linear transformation on the features processed by the ReLU activation function; The calculation formula for the loss function is: ； In the formula, is the single-task loss, is the task weight coefficient, is the mean squared error, is the true value of the i-th task.

5. The moss detection method for GIS equipment based on temperature and humidity prediction and image segmentation according to claim 4, characterized in that, Step 4 includes the following sub-steps: Step 4.1: According to the output result of the temperature and humidity prediction model in Step 3, collect the temperature and humidity data of the substation for the next 7 days; Step 4.2: Construct a temperature and humidity weighted scoring model using the temperature and humidity weighted comprehensive screening method, and calculate the final comprehensive score ; Step 4.2.1: Set the grading rules for temperature T and humidity H, and calculate the temperature score of the substation and humidity score ; Step 4.2.2: Calculate the time weight according to the consecutive days P that meet the moss growth conditions ; Step 4.2.3: Calculate the daily comprehensive score , and calculate the final comprehensive score according to the time weight of the consecutive days P ; ; Step 4.3: According to the final comprehensive score S, set the threshold of moss growth conditions to determine the risk range of moss growth in the substation; Step 4.4: Obtain the screened target substations.

6. The moss detection method for GIS equipment based on temperature and humidity prediction and image segmentation according to claim 5, characterized in that, In Step 4.2.1, the grading and scoring rules for the temperature T and humidity H specifically include: Temperature score and humidity score The calculation formulas are respectively as follows: ； 。 7. The moss detection method for GIS equipment based on temperature and humidity prediction and image segmentation according to claim 6, wherein, In Step 4.2.2, the time weight is calculated by the following formula: 。 8. The moss detection method for GIS equipment based on temperature and humidity prediction and image segmentation according to claim 7, characterized in that, In Step 4.2.3, the final comprehensive score is calculated according to the following formula: ； ； In the formula, is the temperature weight, is the humidity weight.

9. The method for detecting moss on GIS equipment based on temperature and humidity prediction and image segmentation according to claim 8, wherein, In Step 4.3, the threshold of moss growth conditions specifically includes: 。 10. The moss detection method for GIS equipment based on temperature and humidity prediction and image segmentation according to claim 9, characterized in that, Step 5 includes the following sub-steps: Step 5.1: Collect the surface image data of the GIS equipment of the target substation obtained in Step 4 and convert the data from the RGB color space to the HSI color space; Step 5.2: For the HSI color space obtained in Step 5.1, construct a cyclic energy curve CEC on the H component of the HSI color space and calculate the pixel value energy; The formula for calculating the pixel value energy of the cyclic energy curve CEC is: ； In the formula, is the pixel value, is the neighborhood system of the pixel; Step 5.3: Use Tsallis entropy to calculate the segmentation threshold K of the CEC. The calculation formula for Tsallis entropy is: ； In the formula, is the parameter of Tsallis entropy, is the probability distribution of pixels; Step 5.4: Segment the linearized CEC according to the segmentation threshold K in Step 5.3; Step 5.5: Detect the location of moss on the surface of the GIS device according to the segmentation result in Step 5.4.

Images