Rod-plate flame gap DC breakdown voltage prediction method and system based on feature extraction and WOA-WRF

Abstract

A rod-plate flame gap DC breakdown voltage prediction method and system based on feature extraction and WOA-WRF, and the system, belong to the technical field of power system safety monitoring. The present invention aims to solve the problems of low accuracy and weak model generalization ability in the prior art of rod-plate flame gap DC breakdown voltage prediction. The technical solution of the present invention is as follows: first, characteristic factors strongly correlated with the rod-plate flame gap DC breakdown voltage are extracted from test data; then, a whale optimization-weighted random forest algorithm model is constructed, and the weights of the weighted random forest model are adaptively optimized using the whale optimization algorithm; finally, the extracted characteristic factors are input into the optimized weighted random forest model to obtain the predicted value of the rod-plate flame gap DC breakdown voltage. The present invention significantly improves the prediction accuracy by comprehensively considering the static and dynamic characteristics of the flame and combining the whale optimization algorithm, providing strong support for power system safety monitoring.

Description

Technical Field

[0001] The present invention relates to the field of electrical engineering power transmission technology, and in particular to a rod-plate flame gap DC breakdown voltage prediction method and system based on feature extraction and WOA-WRF. Background Art

[0002] The complex operating environment of distribution networks, with densely packed facilities and confined working areas, poses significant safety challenges to on-site workers. During non-stop operations, high-intensity electric fields, transient and steady-state electric shocks, and potential air gap discharges caused by misoperation pose serious threats to worker safety. Therefore, safety precautions during non-stop operations are crucial to ensuring smooth operation.

[0003] In the field of electrical engineering transmission, the safe operation of transmission lines is crucial for ensuring stable grid operation. However, natural disasters such as wildfires pose a serious threat to transmission lines. Under wildfire conditions, factors such as flames bridging transmission lines, high temperatures causing a decrease in air density, and ash or particles filling air gaps can significantly reduce the air insulation level of transmission lines. This can lead to air gap breakdown and cause transmission line tripping accidents. Therefore, accurately predicting the gap breakdown voltage under wildfire conditions is crucial for preventing transmission line tripping and ensuring safe and stable grid operation.

[0004] At present, the prediction methods of gap breakdown voltage are mainly divided into two categories: prediction equation methods based on experience or semi-empirical experience and prediction model methods based on intelligent algorithms.

[0005] Empirical or semi-empirical prediction equations, such as the Peek formula and its modified form, were widely used in the early days. These methods, through statistical analysis of extensive experimental data, establish mathematical relationships between the gap breakdown voltage and factors such as gap distance, voltage waveform, air pressure, and humidity. However, these methods are generally only applicable under specific conditions, and their accuracy is affected by multiple factors, including environmental and flame characteristics. These methods are unable to fully reflect the complex variations in gap breakdown voltage under wildfire conditions.

[0006] With the development of computer technology and artificial intelligence, prediction modeling methods based on intelligent algorithms have gradually emerged. These methods utilize intelligent algorithms such as neural networks, support vector machines, and random forests to train and learn from large amounts of experimental data to establish prediction models for gap breakdown voltage. Among them, the random forest algorithm, as an ensemble learning method, improves the model's prediction accuracy and stability by constructing multiple decision trees. It has been widely used in gap breakdown voltage prediction. However, existing random forest-based prediction models still have some problems. For one thing, the model is easily interfered with by redundant features or noisy data during training, resulting in reduced prediction accuracy. For another, the model's weight assignment is often based on experience or default settings, lacking an adaptive optimization mechanism, making it difficult to adapt to the prediction needs under different operating conditions.

[0007] For example, CN103678941B discloses a method for predicting the breakdown voltage of an electrode air gap. First, the breakdown voltage values ​​of typical electrode air gaps with different structures are measured, and the withstand voltage range and breakdown voltage range are defined. Then, the electric field of the typical electrode gaps with different structures is calculated and the electric field characteristics are extracted to construct a training sample set. Then, a breakdown voltage prediction model is constructed based on the training sample set. The breakdown voltage prediction model takes the electric field characteristics as input and the withstand voltage range and breakdown voltage range as output. Finally, the breakdown voltage of the electrode air gap is predicted using the breakdown voltage prediction model. However, the breakdown voltage of the electrode air gap is not predicted based on the breakdown voltage prediction model. The above method targets air gaps and does not consider the situation of vegetation flame gaps. Vegetation combustion not only provides a large number of charged particles for discharge, but also affects the electric field distribution of the gap; DC voltage affects the movement of charged particles, and then affects combustion and spatial temperature distribution, which will have a great impact on the development process of gap breakdown, and its characteristic quantities need to be selected; in addition, some prediction models proposed in related patents are not suitable for small samples, and some are highly dependent on kernel functions and penalty coefficients, and are prone to overfitting; it is necessary to propose targeted intelligent prediction algorithms for the data characteristics of flame gaps with multiple features, nonlinearity, and small samples.

[0008] To overcome the limitations of existing technologies and improve the accuracy and adaptability of gap breakdown voltage prediction, this paper proposes a method for predicting the DC breakdown voltage of a rod-plate flame gap based on feature extraction and the Whale Optimization-Weighted Random Forest (WOA-WRF) algorithm. This method constructs a prediction model by extracting characteristic factors strongly correlated with the breakdown voltage, such as combustion characteristics, spatial temperature, leakage current, and electric field characteristics. Furthermore, the Whale Optimization algorithm is used to adaptively optimize the weights of the weighted random forest model to improve the model's prediction accuracy and adaptability. This innovative method is expected to provide more reliable technical support for the safe operation of transmission lines under wildfire conditions. Summary of the Invention

[0009] The technical problem to be solved by the present invention is to provide a rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF, so as to solve the problem of inaccurate prediction of the DC breakdown voltage of the rod-plate flame gap of the transmission line in the field of electrical engineering transmission, especially under wildfire conditions.

[0010] In order to solve the above technical problems, the technical solution adopted by the present invention is: a rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF, comprising the following steps:

[0011] Step 1: Build a test platform and conduct a rod-plate gap breakdown test under flame conditions. Obtain the breakdown voltage under typical woodpile sizes, gap distances, and different positive and negative polarity characteristics, and record the corresponding leakage current, environmental factors, and other characteristics.

[0012] Step 2: Establish a gap breakdown simulation. By using a multi-physics field coupling method, establish a gap breakdown simulation corresponding to the experiment in Step 1 to obtain characteristic parameters such as space temperature and electric field strength that are difficult to obtain directly through experiments.

[0013] Step 3: Eliminate redundant features, quantify the influence of different feature parameters on the breakdown voltage, and eliminate redundant features that are irrelevant or weakly correlated with the breakdown voltage;

[0014] Step 4: Feature extraction: extract characteristic factors that are strongly correlated with the DC breakdown voltage of the rod-plate flame gap from the test data;

[0015] Step 5: Build a whale optimization-weighted random forest algorithm model, use the whale optimization algorithm to adaptively optimize the weights of the weighted random forest model, and obtain the optimized weighted random forest model;

[0016] Step 6: Application of the prediction model: input the extracted characteristic factors into the optimized weighted random forest model to obtain the predicted value of the DC breakdown voltage of the rod-plate flame gap.

[0017] In the preferred solution, the specific implementation process of Step 2 is that the main combustible component during the combustion of the woodpile is cellulose, and after the woodpile is burned, a large amount of hydrocarbons are generated by reaction. The pyrolysis chemical reaction process generates a large amount of heat. Under the action of high heat, the hydrocarbons undergo ionization reactions with other substances in the air to generate a large number of electrons and positively charged particles. The distance between the rod electrode and the woodpile, the radius and length of the rod electrode, and the radius and height of the woodpile are set, and the model geometric structure simulation is performed to obtain the corresponding electric field distribution and spatial temperature through simulation.

[0018] In the preferred solution, the specific method of Step 3 is to analyze the influence of the input vector characteristics on the prediction results, use the correlation coefficient method to remove the characteristic parameters that are irrelevant or weakly correlated with the breakdown voltage, calculate the Pearson coefficient of each characteristic parameter for the breakdown voltage according to the following formula, and remove redundant features with low correlation to improve the prediction accuracy.

[0019] Furthermore, the characteristic factors in Step 4 include combustion characteristics, space temperature, leakage current and electric field characteristics, and the combustion characteristics include flame height, flame width and flame color.

[0020] Furthermore, the implementation method of Step 4 is to consider the flame bridging situation, electric field distribution characteristics, discharge and development channel factors under flame conditions, extract characteristic parameters related to combustion characteristics, spatial temperature, leakage current, electric field characteristics and breakdown voltage, and define the calculation method of related characteristics. The specific implementation steps include:

[0021] Step 4.1: Select the spatial temperature, leakage current and other characteristics on the shortest path to characterize the flame gap breakdown and extract the characteristic parameters;

[0022] Step 4.2: The input characteristic values ​​are all taken within a flame oscillation cycle. The specific definitions of the flame environment characteristic values ​​include: flame root temperature , maximum temperature and average temperature ; Temperature near the rod electrode , maximum temperature and average temperature ; Rod-plate gap distance , Maximum flame height ; Leakage current is , the average value of the leakage current during the stable phase , maximum value , standard deviation , leakage current ripple factor ; Field strength around the electrode , its maximum field strength is , average field strength : discharge channel field strength , its maximum field strength is , average field strength , energy density .

[0023] In the preferred solution, in Step 5, the specific process of the whale optimization algorithm includes: initializing the whale population, the whale search process and judging the termination condition, and finally outputting the optimal solution as the weight of the weighted random forest model. The specific implementation steps of Step 5 include:

[0024] Step 5.1: Preprocess the sample data, including feature parameter dimensionality reduction, data normalization, and sample sorting to improve the generalization ability and accuracy of the model;

[0025] Step 5.2: Perform sample extraction, parameter selection and model training, and use Bootstrap resampling to training samples ( ) times of random sampling with replacement, forming Each subset of samples has different characteristics, and a decision tree is constructed based on the mean square error. Select from the features ( ) features to train binary tree CART, in the initial stage, Pick , Pick The model prediction result is the mean of the voltage predicted by all decision trees. The out-of-bag data (OOB) that has not been extracted is used as input. The OOB error is calculated to test the model performance. The number of decision trees or the number of feature parameters is changed multiple times until the OOB error no longer changes significantly.

[0026] Step 5.3: The random forest algorithm uses random sampling and random feature selection to construct a weighted decision tree based on the correlation of feature parameters. The Whale Algorithm (WOA) is then used to adaptively adjust the weight of each decision tree, continuously changing the whale's position, i.e., the decision tree weight. The algorithm then approaches the prey, i.e., the optimal solution, through searching, encircling, and predation behaviors until the model's fitness shows no significant improvement. The whale moves forward in a spiral, continuously updating its position, and iterates until it captures its prey.

[0027] Step 5.4: When establishing the prediction model, in order to improve the prediction accuracy and operation speed of the model, the feature data is processed, the Pearson correlation coefficient method is selected for feature parameter selection, and the selected feature parameters are normalized;

[0028] Step 5.5: Select a typical performance evaluation indicator in the random forest regression algorithm, using the root mean square error , coefficient of determination , mean absolute percentage error Perform an overall characterization and evaluation of the model performance.

[0029] In a preferred embodiment, the application of the prediction model in Step 6 also includes verification of the accuracy of the prediction model by comparing the predicted values ​​with the actual measured values ​​to evaluate the performance of the prediction model.

[0030] In a preferred solution, the method further considers the influence of the dynamic characteristics of the flame on the breakdown voltage, and adds the extraction of the dynamic characteristics of the flame, namely the flame jumping frequency and the flame spreading speed, in the feature extraction Step 1.

[0031] In a preferred solution, the flame dynamic characteristics are input as additional characteristic factors together with the static characteristic factors into the whale optimization-weighted random forest algorithm model for prediction.

[0032] A system for implementing any of the above-mentioned rod-plate flame gap DC breakdown voltage prediction methods based on feature extraction and WOA-WRF, comprising:

[0033] Feature extraction module, used to extract characteristic factors that are strongly correlated with the DC breakdown voltage of the rod-plate flame gap from the test data;

[0034] The model building module is used to build a whale optimization-weighted random forest algorithm model and use the whale optimization algorithm to adaptively optimize the weights of the weighted random forest model;

[0035] A prediction module is used to input the extracted characteristic factors into the optimized weighted random forest model to obtain the predicted value of the DC breakdown voltage of the rod-plate flame gap;

[0036] The verification module is used to verify the accuracy of the prediction model and evaluate the performance of the prediction model by comparing the predicted values ​​with the actual measured values.

[0037] The rod-plate flame gap DC breakdown voltage prediction method and system based on feature extraction and WOA-WRF provided by the present invention have the following beneficial effects:

[0038] 1. The present invention solves the problem of inaccurate prediction of DC breakdown voltage between the flame gap of transmission line rods and plates in the field of electrical engineering transmission, especially under wildfire conditions;

[0039] 2. The present invention surpasses existing technologies in gap breakdown voltage prediction. By extracting characteristic factors, it not only considers traditional factors such as space temperature, leakage current, and electric field characteristics, but also innovatively incorporates combustion characteristics, including combustion rate, flame temperature, and their interaction with the electric field. This measure makes the prediction model more comprehensive and accurate, avoids the limitations of single-variable prediction, provides a strong guarantee for the safe operation of transmission lines, and more comprehensively characterizes the flame gap breakdown of transmission lines under different operating conditions, avoiding the limitations of single-variable or small-variable predictions.

[0040] 3. The prediction method of the present invention innovatively considers the multi-feature, nonlinear, and small sample data characteristics of gap breakdown under flame conditions, and proposes a prediction method based on whale optimization-weighted random forest, which achieves accurate processing of multi-feature and nonlinear data;

[0041] 4. Based on the characteristics of random forests, the present invention analyzes and quantifies the correlation between characteristic parameters and breakdown voltage, and predicts breakdown voltage using weighted decision trees with differences to avoid overfitting of the prediction model. The weights are adaptively optimized through the whale algorithm to further improve the accuracy of model prediction.

[0042] 5. The present invention combines the advantages of random forest and whale algorithms. The breakdown voltage prediction model has high accuracy and good robustness. The model can maintain stable performance under different working conditions and has strong generalization ability. The model can be applied to data from a variety of different situations.

[0043] 6. The feature extraction adopted by the present invention extracts multiple characteristic parameters that are strongly correlated with the breakdown voltage, which is more comprehensive than the previous prediction methods based on a single variable or a few variables;

[0044] 7. The advanced prediction model of the present invention combines weighted random forest with whale algorithm to establish a breakdown voltage prediction model with high accuracy, good robustness and strong generalization ability, and realizes accurate processing of multi-feature and nonlinear data;

[0045] 8. The present invention combines experiments with simulations. By building a test platform and establishing a simulation model, rich test and simulation data are obtained, providing a reliable basis for model training and verification.

[0046] 9. The present invention effectively solves the prediction problem under small sample data and realizes the accurate prediction of the DC breakdown voltage of the rod-plate flame gap;

[0047] 10. The present invention has a wide range of applications and can be used to predict transmission line gap breakdown under different working conditions, thus expanding the scope of application of the prediction model.

[0048] 11. The present invention provides a reliable prediction basis for adjusting the transmission line operation strategy, which helps to protect the external insulation of the transmission line and reduce tripping accidents;

[0049] 12. The present invention has significant innovation and application value in the field of electrical engineering transmission, especially in the prediction of the DC breakdown voltage of the flame gap between the rod and plate of the transmission line under wildfire conditions, providing a more accurate and reliable solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0051] Figure 1 This is a schematic diagram of the characteristic value area of ​​the present invention;

[0052] Figure 2 This is a breakdown voltage prediction flow chart of the present invention;

[0053] Figure 3 This is a schematic diagram of the gap breakdown test arrangement of the present invention;

[0054] Figure 4 Schematic diagram of the breakdown voltage test values ​​under different working conditions of the present invention;

[0055] Figure 5 Schematic diagram of the geometric structure and value range of the true model of the present invention;

[0056] Figure 6 Schematic diagram of electric field distribution under different working conditions of 50cm gap of the present invention;

[0057] Figure 7 Schematic diagram of the temperature near the electrode under different voltages of the present invention;

[0058] Figure 8 Schematic diagram comparing the long-distance gap breakdown voltage prediction results of the present invention. DETAILED DESCRIPTION

[0059] The technical solutions of the present invention are further described below with reference to the accompanying drawings and embodiments:

[0060] Example 1

[0061] A rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF (Whale Optimization Algorithm-Weather Research and Forecasting) includes the following steps:

[0062] Step 1: Build a test platform and conduct a rod-plate gap breakdown test under flame conditions. Obtain the breakdown voltage under typical woodpile sizes, gap distances, and different positive and negative polarity characteristics, and record the corresponding leakage current, environmental factors, and other characteristics.

[0063] Step 2: Establish a gap breakdown simulation. By using a multi-physics field coupling method, establish a gap breakdown simulation corresponding to the experiment in Step 1 to obtain characteristic parameters such as space temperature and electric field strength that are difficult to obtain directly through experiments.

[0064] Step 3: Eliminate redundant features, quantify the influence of different feature parameters on the breakdown voltage, and eliminate redundant features that are irrelevant or weakly correlated with the breakdown voltage;

[0065] Step 4: Feature extraction: extract characteristic factors that are strongly correlated with the DC breakdown voltage of the rod-plate flame gap from the test data;

[0066] Step 5: Build a whale optimization-weighted random forest algorithm model, use the whale optimization algorithm to adaptively optimize the weights of the weighted random forest model, and obtain the optimized weighted random forest model;

[0067] Step 6: Application of the prediction model: input the extracted characteristic factors into the optimized weighted random forest model to obtain the predicted value of the DC breakdown voltage of the rod-plate flame gap.

[0068] In this embodiment, the specific implementation steps in Step 1 include:

[0069] Step 1.1: Build the test platform, place the woodpile in the middle of the mesh electrode and vertically align it with the rod electrode above, and connect the signal acquisition system to the circuit;

[0070] Step 1.2: Use a spray bottle to evenly spray industrial alcohol on the woodpile. After igniting the woodpile, turn on the signal acquisition system and video recorder. After the woodpile is burning stably, use the direct lift method to pressurize the gap until it breaks through.

[0071] Step 1.3: After the flame is extinguished, change the voltage polarity and gap distance, rearrange the woodpile, and repeat the test according to the above steps to obtain the breakdown voltage and leakage current when the flame bridges the rod-plate gap.

[0072] Furthermore, the specific implementation process of Step 2 is that the main combustible component in the woodpile combustion process is cellulose, and the simplified molecular formula is , then the combustion reaction equation when the pine wood pile undergoes pyrolysis is:

[0073] (1)

[0074] After the woodpile is burned, a large amount of hydrocarbons are generated. The pyrolysis chemical reaction process generates a large amount of heat. Under the action of high heat, the hydrocarbons react with other substances in the air to produce a large number of electrons and positively charged particles. The main ion reactions are as follows:

[0075] (2)

[0076] (3)

[0077] The distance between the rod electrode and the woodpile, the radius and length of the rod electrode, and the radius and height of the woodpile are set, and the model geometry simulation is performed to obtain the corresponding electric field distribution and spatial temperature through simulation.

[0078] Furthermore, the specific method of Step 3 is to analyze the influence of the input vector features on the prediction results, use the correlation coefficient method to remove the feature parameters that are irrelevant or weakly correlated with the breakdown voltage, calculate the Pearson coefficient of each feature parameter for the breakdown voltage according to the following formula, and remove redundant features with low correlation to improve the prediction accuracy:

[0079] (4)

[0080] Where, and Represents a characteristic parameter and breakdown voltage No. values; Characteristic parameter The average value of Breakdown voltage The average value of .

[0081] Furthermore, the characteristic factors in Step 4 include combustion characteristics, space temperature, leakage current and electric field characteristics, and the combustion characteristics include flame height, flame width and flame color.

[0082] Furthermore, the implementation method of Step 4 is to consider the flame bridging situation, electric field distribution characteristics, discharge and development channel factors under flame conditions, extract characteristic parameters related to combustion characteristics, spatial temperature, leakage current, electric field characteristics and breakdown voltage, and define the calculation method of related characteristics. The specific implementation steps include:

[0083] Step 4.1: Select the spatial temperature, leakage current and other characteristics on the shortest path to characterize the flame gap breakdown and extract the characteristic parameters;

[0084] Step 4.2: The input feature quantities are all taken from one flame oscillation cycle.

[0085] Furthermore, the specific definition of the flame environment characteristic value in Step 4.2 is as follows:

[0086] Step 4.2.1: Flame root temperature , maximum temperature and average temperature :

[0087] (5)

[0088] (6)

[0089] Where, For the root of the flame The temperature of the unit, is the total number of units;

[0090] Step 4.2.2: Temperature near the rod electrode , maximum temperature and average temperature :

[0091] (7)

[0092] (8)

[0093] Where, The rod electrode The temperature of the unit, is the total number of units;

[0094] Step 4.2.3: Rod-plate gap distance , Maximum flame height Respectively expressed as:

[0095] (9)

[0096] (10)

[0097] Where, 、 are the heights of the flame continuous zone and the flame gap zone, respectively. is the coefficient of the flame bridge rod electrode, ranging from 0 to 1;

[0098] Step 4.2.4: Leakage current is , respectively take the average value of the leakage current before the gap breakdown and the stable stage , maximum value , standard deviation , leakage current ripple factor It is expressed as follows:

[0099] (11)

[0100] Where, is the effective value of the leakage current of this group;

[0101] Step 4.2.5: Field strength around the electrode , its maximum field strength is , average field strength :

[0102] (12)

[0103] (13)

[0104] Where, The electrode around The temperature of the unit, is the total number of units;

[0105] Step 4.2.6: Discharge channel field strength , its maximum field strength is , average field strength , energy density :

[0106] (14)

[0107] (15)

[0108] (16)

[0109] Where, The electrode around The temperature of the unit, is the total number of units; 、 Respectively The energy and volume of a unit.

[0110] Furthermore, in Step 5, the specific process of the whale optimization algorithm includes: initializing the whale population, the whale search process, and judging the termination condition, and finally outputting the optimal solution as the weight of the weighted random forest model. The specific implementation steps of Step 5 include:

[0111] Step 5.1: Preprocess the sample data, including feature parameter dimensionality reduction, data normalization, and sample sorting to improve the generalization ability and accuracy of the model;

[0112] Step 5.2: Perform sample extraction, parameter selection and model training, and use Bootstrap resampling to training samples ( ) times of random sampling with replacement, forming Each subset of samples has different characteristics, and a decision tree is constructed based on the mean square error. Select from the features ( ) feature training binary tree (CART, Classification and Regression Tree), in the initial stage, Pick , Pick The model prediction result is the mean of the voltage predicted by all decision trees. The out-of-bag data (OOB) that has not been extracted is used as input. The OOB error is calculated to test the model performance. The number of decision trees or the number of feature parameters is changed multiple times until the OOB error no longer changes significantly.

[0113] Step 5.3: Random Forest uses random sampling and random feature selection to construct a weighted decision tree using the correlation of feature parameters. The Whale Optimization Algorithm (WOA) is then used to adaptively adjust the weight of each decision tree, continuously changing the position of the whale, i.e., the weight of the decision tree. The algorithm then approaches the prey through search, encirclement, and predation, until the model's fitness does not improve significantly. The calculation process is as follows:

[0114] (17)

[0115] (18)

[0116] Where, For decision tree The weight of is the number of decision trees; For decision tree The sum of the correlation degrees of all features can be expressed as , is the feature in the decision tree The degree of correlation, is the number of features; is the predicted breakdown voltage, For decision tree Predicted breakdown voltage;

[0117] (19)

[0118] Where, is the number of iterations; is the coefficient, expressed as , A random number between 0 and 1; The initial value of is 2 and it is a random number that decreases to 0 during the iteration process; is the current position of the whale; The position of a random whale; is the distance between the whale and its prey, expressed as ;

[0119] (20)

[0120] Where, is the current prey position;

[0121] Whales travel in spirals:

[0122] (twenty one)

[0123] Where, Used to define the logarithmic spiral shape when the whale updates its position. is a constant; A random number between -1 and 1;

[0124] The whale's position is continuously updated, and the cycle iterates until the prey is captured.

[0125] Step 5.4: When establishing a prediction model, in order to improve the prediction accuracy and operation speed of the model, the feature data is processed and the Pearson correlation coefficient method is selected to select the feature parameters. The selected feature parameters are normalized. The calculation formula is as follows:

[0126] (twenty two)

[0127] Where, is an input parameter; and are the maximum and minimum values ​​of the parameter respectively; is the normalized value;

[0128] Step 5.5: Select a typical performance evaluation indicator in the random forest regression algorithm, using the root mean square error , coefficient of determination , mean absolute percentage error Overall characterization and evaluation of model performance:

[0129] (twenty three)

[0130] (twenty four)

[0131] (25)

[0132] Where, For the The actual breakdown voltage of the samples; For the The predicted breakdown voltage of samples; is the total number of samples.

[0133] Furthermore, the application of the prediction model in Step 6 also includes verification of the accuracy of the prediction model by comparing the predicted values ​​with the actual measured values ​​to evaluate the performance of the prediction model.

[0134] Furthermore, the method further considers the influence of the dynamic characteristics of the flame on the breakdown voltage, and adds the extraction of the dynamic characteristics of the flame, namely the flame jumping frequency and the flame spreading speed, in the feature extraction Step 1.

[0135] Furthermore, the flame dynamic characteristics are input as additional characteristic factors together with the static characteristic factors into the whale optimization-weighted random forest algorithm model for prediction.

[0136] Example 2

[0137] In another preferred embodiment, based on the above embodiment 1, a system for implementing any of the above-mentioned rod-plate flame gap DC breakdown voltage prediction methods based on feature extraction and WOA-WRF includes:

[0138] Feature extraction module, used to extract characteristic factors that are strongly correlated with the DC breakdown voltage of the rod-plate flame gap from the test data;

[0139] The model building module is used to build a whale optimization-weighted random forest algorithm model and use the whale optimization algorithm to adaptively optimize the weights of the weighted random forest model;

[0140] A prediction module is used to input the extracted characteristic factors into the optimized weighted random forest model to obtain the predicted value of the DC breakdown voltage of the rod-plate flame gap;

[0141] The verification module is used to verify the accuracy of the prediction model and evaluate the performance of the prediction model by comparing the predicted values ​​with the actual measured values.

[0142] Example 3

[0143] In another preferred embodiment, based on the above embodiments 1 and 2, this embodiment elaborates on the rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF.

[0144] A rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF first considers the flame bridging situation and gap discharge development process under wildfire conditions to extract characteristic parameters that affect the breakdown voltage. Secondly, considering the data characteristics of the gap breakdown test and the nonlinear correlation between the characteristic parameters, a prediction model based on whale optimization-weighted random forest is established. Then, a gap breakdown test is carried out under flame conditions to obtain the gap breakdown voltage and the corresponding characteristic parameters. A corresponding gap breakdown multi-physics field coupling simulation model is then established to obtain accurate characteristic parameters such as temperature and electric field. Finally, the prediction model is trained and the effectiveness and generalization ability of the model are verified. The specific steps are as follows:

[0145] S1: Considering factors such as flame bridging, electric field distribution characteristics, discharge and development channels under wildfire conditions, extract characteristic parameters related to breakdown voltage, such as combustion characteristics, spatial temperature, leakage current, and electric field characteristics, and define the calculation method of related characteristics. The specific steps are as follows:

[0146] S1.1: Feature value area such as Figure 1 As shown in the figure, in the absence of wind, the flame shape of the burning woodpile is distributed upward in a cone shape. Affected by the particle concentration, the discharge arc breaks through the air gap along the flame core (or fire plume, smoke).

[0147] Existing research shows that the discharge channel has a strong correlation with the electric field distribution in the shortest path region between electrodes and the development of gap discharge; the breakdown voltage under flame conditions has a significant correlation with the spatial temperature and charged particles; the spatial temperature and charged particles in different regions show a polarity effect distribution under the action of the voltage applied by the electrodes; the concentration of spatial charged particles in the flame has a strong correlation with the leakage current. Therefore, the spatial temperature and leakage current on the shortest path are used to characterize the flame gap breakdown. The extracted characteristic parameters are shown in Table 1:

[0148]

[0149] S1.2: The input characteristic values ​​are all taken within a flame oscillation cycle. The specific definition of the flame environment characteristic value is as follows:

[0150] (1) Flame root temperature , maximum temperature and average temperature :

[0151] (5)

[0152] (6)

[0153] (2) Temperature near the rod electrode , maximum temperature and average temperature :

[0154] (7)

[0155] (8)

[0156] Where, The rod electrode The temperature of the unit, is the total number of units;

[0157] (3) Rod-plate gap distance , Maximum flame height Respectively expressed as:

[0158] (9)

[0159] (10)

[0160] (4) Leakage current is , respectively take the average value of the leakage current before the gap breakdown and the stable stage , maximum value , standard deviation , leakage current ripple factor It is expressed as follows:

[0161] (11)

[0162] (5) Field strength around the electrode , its maximum field strength is , average field strength :

[0163] (12)

[0164] (13)

[0165] (6) Discharge channel field strength , its maximum field strength is , average field strength , energy density :

[0166] (14)

[0167] (15)

[0168] (16).

[0169] S2: Considering the data characteristics of gap breakdown, such as multiple features, nonlinearity, and small samples, a breakdown voltage prediction model based on weighted random forest-whale adaptive optimization is established. The specific steps are as follows:

[0170] S2.1: Random forest is an integrated learning model that randomly extracts breakdown voltage samples and randomly extracts split features to form different decision trees to build a model to improve the accuracy, robustness and generalization ability of model prediction. When using random forest for regression prediction, the prediction result of breakdown voltage is the average of the prediction results of each decision tree. The modeling process is as follows: Figure 2 As shown:

[0171] First, the sample data is preprocessed, including feature parameter dimension reduction, data normalization, and sample sorting, to improve the generalization ability and accuracy of the model; then, sample extraction, parameter selection and model training are carried out, and Bootstrap resampling is used to training samples ( ) times of random sampling with replacement, forming The method is to construct a decision tree based on the mean square error for each subset of samples, and randomly select the subset from the sample group in the split. Select from the features ( ) features to train binary tree CART, in the initial stage, Pick , Pick The model prediction result is the mean of the voltage predicted by all decision trees. The out-of-bag data OOB that has not been extracted is used as input. The OOB error is calculated to test the model performance. The number of decision trees or the number of feature parameters is changed multiple times until the OOB error no longer changes significantly.

[0172] S2.2: Random forest ensures diversity among decision trees by random sampling and random selection of features to avoid overfitting. In the conventional calculation process, the weights of the decision trees are equal. In order to improve the accuracy of the model, the correlation degree of the feature parameters is used to construct a weighted decision tree, and the weight of each decision tree is adaptively adjusted in combination with the whale algorithm WOA. Initially, the weight of each decision tree is determined by the selected features, and the model prediction error is calculated iteratively. The whale position (i.e., the decision tree weight) is continuously changed, and then the prey (i.e., the optimal solution) is approached through search, encirclement, and predation behaviors until the fitness of the model is not significantly improved. The calculation process is as shown in formulas (17) to (20);

[0173] Whales travel in spirals:

[0174] (twenty one)

[0175] Combining equations (17) to (19), the whale's position is continuously updated and the loop is iterated until the prey is captured. Finally, the coefficient of determination, root mean square error, and mean absolute percentage error are used to characterize and evaluate the model performance.

[0176] S2.3: When building a prediction model, feature data needs to be processed to improve the prediction accuracy and computing speed of the model:

[0177] (1) Feature parameter dimensionality reduction

[0178] Feature parameters are an important option in model prediction. Too many non-correlated or weakly correlated feature parameters will reduce the accuracy of random forest regression prediction. Therefore, the Pearson correlation coefficient method is used to select feature parameters. The calculation formula is as follows:

[0179] (4)

[0180] (2) Feature parameter normalization

[0181] In order to reduce the impact of different attribute feature parameters on model accuracy and convergence speed, the filtered feature parameters are normalized. The calculation formula is as follows:

[0182] (twenty two).

[0183] S2.4: Select typical performance evaluation indicators in the random forest regression algorithm to quantify the robustness and generalization ability of the model, such as root mean square error , coefficient of determination , mean absolute percentage error The overall performance of the model is characterized and evaluated using formulas (23) to (25).

[0184] S3: Build a test platform and conduct a rod-plate gap breakdown test under flame conditions. Obtain the breakdown voltage under typical woodpile sizes, gap distances, and different positive and negative polarity characteristics, and record the corresponding leakage current, environmental factors, and other characteristics. The specific steps are as follows:

[0185] S3.1: Test apparatus for the rod-plate gap breakdown test under burning wood pile conditions Figure 3 The test device consists of a DC power supply, a protective resistor, a voltage divider, a rod electrode, a wood pile, a metal bracket, and a breakdown voltage acquisition device. The DC power supply is a 200kV DC power generator, and the breakdown voltage is measured using a 5000:1 resistor-capacitor voltage divider connected to the acquisition device.

[0186] The arrangement of wood piles is simple in structure and has good repeatability. Using wood pile fire as the fire source, the effects of different flame intensities on the gap breakdown voltage were compared. The experimental wood piles were approximately circular, consisting of regular pine wood strips measuring 2 cm × 2 cm × 7 cm.

[0187] During the test, pine wood strips were arranged into approximately circular piles with diameters of 20cm, 26cm, 32cm, and 62cm, and a height of 6cm. Breakdown tests were conducted at different gap distances. During the pressurization process, the applied voltage and waveform were recorded simultaneously, and each test was recorded with a camera. The specific test steps are as follows:

[0188] S3.1.1: Press Figure 1Build a test platform, place the woodpile in the middle of the mesh electrode and vertically align it with the rod electrode above, and connect the signal acquisition system to the loop;

[0189] S3.1.2: Use a spray bottle to evenly spray denatured alcohol onto the woodpile. After igniting the woodpile, activate the signal acquisition system and video recorder. Once the woodpile is burning steadily, apply pressure to the gap using the direct voltage method (2-3 kV / s) until breakdown occurs.

[0190] S3.1.3: After the flame is extinguished, change the voltage polarity and gap distance, rearrange the woodpile, and repeat the test according to the above steps to obtain the breakdown voltage and leakage current when the flame bridges the rod-plate gap;

[0191] During the test, each voltage polarity and gap distance was repeated 6 times. All tests were carried out in a semi-enclosed space to reduce the impact of external wind on the test.

[0192] S3.2: Under woodpile burning conditions, the flame replaces all or part of the air gap. The breakdown voltage of the rod-plate gap will be significantly reduced compared to a pure air gap. When the flame gap breaks down, a bright arc develops along the bright area in the center of the flame and penetrates the entire gap. Figure 4 The breakdown voltage values ​​for different woodpile sizes and different flame gaps are shown. It can be seen that as the rod-plate gap increases, the breakdown voltage gradually increases, and the gradient of the breakdown voltage also increases, with the tail of the curve showing a slight upward trend.

[0193] At the same gap distance, the gap breakdown voltage decreases as the size of the woodpile increases. Analysis shows that the flame radius of a large woodpile is larger and the concentration of charged particles in the flame increases, making it easier to form a stable arc channel, resulting in a lower breakdown voltage. Under the same operating conditions, the breakdown voltage of negative polarity is higher than that of positive polarity. Analysis shows that the mass of electrons is much smaller than that of positive ions. Under the action of the electric field, a large number of electrons quickly move toward the positive electrode, making it easier to form a stable conductive channel.

[0194] S4: Using the multi-physics field coupling method, establish a gap breakdown simulation corresponding to the above test to obtain characteristic parameters such as spatial temperature and electric field strength that are difficult to obtain directly through experiments. The specific steps are as follows:

[0195] S4.1: According to research by domestic and foreign scholars, the main combustible component in the combustion of wood piles is cellulose, and its simplified molecular formula is , then the combustion reaction equation when the pine wood pile undergoes pyrolysis is:

[0196] (1)

[0197] After the woodpile is burned, a large amount of hydrocarbons are generated. The pyrolysis chemical reaction process generates a large amount of heat. Under the action of high heat, the hydrocarbons react with other substances in the air to produce a large number of electrons and positively charged particles. The main ion reactions are as follows:

[0198] (2)

[0199] (3)

[0200] Distance between rod electrode and wood pile According to the different electrode heights under specific test conditions, the rod electrode radius is 1.5 cm, the length is 9.5 cm, and the air bag with a radius of 500 cm and a height of 2500 cm is set as the outer boundary; the simulation model geometry is as follows Figure 5 shown.

[0201] S4.2: The electric field distribution under different working conditions is as follows Figure 6 As shown in the figure, under flame conditions, when a 45kV or 55kV DC positive voltage is applied to the rod electrode, the maximum field strength is 1.73×10 4 kV / m, 2.07×10 4 kV / m; when the rod electrode is applied with -45kV and -55kV DC negative voltage, the maximum field strength is 1.58×10 4 kV / m, 1.91×10 4 kV / m;

[0202] The electric field strength near the positive electrode is significantly greater than that near the negative electrode. Analysis shows that under flame conditions, there are a large number of free positive ions, negative ions and electrons in the air gap, which will move toward the opposite electrode under the action of Coulomb force. A large number of negative ions and electrons will gather near the rod electrode with positive voltage applied, while a large number of positive ions will gather near the rod electrode with negative voltage applied.

[0203] Depend on Figure 7 As shown in the figure, under the action of voltage, the temperature near the electrode is higher. Compared with the temperature rise of 22℃~113℃ when no voltage is applied, the positive and negative voltages have different effects on the temperature near the electrode. As the voltage increases, the temperature near the electrode gradually increases. The temperature near the electrode shows an obvious polarity effect. Under different voltage amplitudes, the temperature near the electrode under positive polarity voltage rises by 12℃~61℃ compared with the temperature under negative polarity voltage.

[0204] The analysis shows that under the action of the electric field force, the electric field force in the area near the electrode increases. In this area, the charged particles are simultaneously affected by the fluid uplift and the electric field force, which increases the probability of particle ionization and collision ionization, accelerates the process of chemical reaction in this area, and the temperature in the area near the electrode rises. In the subsequent determination of the spatial temperature characteristic parameters, the area near the electrode and the flame body of the discharge channel are mainly considered.

[0205] S5: Quantify the influence of different characteristic parameters on the breakdown voltage and eliminate redundant features that are irrelevant or weakly correlated with the breakdown voltage;

[0206] In order to analyze the influence of input vector features on the prediction results, the correlation coefficient method is used to remove the characteristic parameters that are irrelevant or weakly correlated with the breakdown voltage. The Pearson coefficient of each characteristic parameter for the breakdown voltage is calculated according to formula (4), as shown in Table 2; the maximum temperature at the root of the flame is removed. ,average value , Maximum leakage current , leakage current standard deviation , discharge channel energy density Several redundant features with low relevance:

[0207]

[0208] S6: Based on the prediction model established in step 2 and the sample data obtained in steps 3 and 4, train the prediction model and verify the effectiveness of the model:

[0209] S6.1: Arrange the breakdown voltage data obtained from the experiment, disrupt the order of the samples, and divide them into training samples and test samples. Use the feature parameters of the corresponding samples after dimensionality reduction and all feature parameters without dimensionality reduction as inputs to the model, and calculate the evaluation indicators of the model:

[0210]

[0211] The evaluation indicators based on the whale algorithm-weighted random forest model are shown in Table 3. Under positive polarity voltage, the root mean square error of the flame gap breakdown voltage of the three different sizes of wood piles is reduced by 1.458, 1.312, and 1.019 respectively after feature dimensionality reduction; the determination coefficient is increased by 0.042, 0.047, and 0.048, and the average absolute percentage error is reduced by 3.5%, 4.4%, and 3.6% respectively; under negative polarity voltage, the root mean square error of the flame gap breakdown voltage of the three different sizes of wood piles is reduced by 1.284, 1.311, and 1.623 respectively after feature dimensionality reduction; the determination coefficient is increased by 0.031, 0.044, and 0.042, and the average absolute percentage error is reduced by 3.8%, 4.0%, and 4.1% respectively; the error of the predicted value after dimensionality reduction is smaller and closer to the experimental breakdown voltage value. The results show that using the correlation coefficient method for feature dimensionality reduction and removing features that are irrelevant or weakly correlated with the breakdown voltage can help improve the model prediction accuracy.

[0212] S6.2: In order to explore the prediction effect of different prediction methods on the breakdown voltage under longer gap flame conditions, the long gap breakdown voltage under flame conditions was predicted with reference to the two methods of regression analysis under flame and SVM model prediction; the above two methods were compared with the breakdown voltage prediction method based on whale algorithm-weighted random forest proposed in this invention; all methods used the above-mentioned experimental 30~60cm small gap breakdown voltage data as training samples, and 100cm~140cm breakdown voltage data as test samples.

[0213] The comparison results are as follows Figure 8 As shown in the figure, the average errors of the positive and negative polarity breakdown voltage prediction results using linear regression analysis are 5.1% and 5.2%, respectively. The average errors of the SVM model for long-gap breakdown voltage prediction are 5.5% and 4.9%, respectively. The average errors of the WOA-WRF model for positive and negative polarity breakdown voltage prediction results are 2.1% and 2.4%, respectively. It can be seen that the linear regression and SVM models have poor generalization capabilities. In contrast, the WOA-WRF model has relatively low prediction errors and performs well even for longer gaps. These comparative results verify the effectiveness of the WOA-WRF-based breakdown voltage prediction model and its applicability to long-gap breakdown prediction.

[0214] Example 4

[0215] In another preferred embodiment, based on Example 3, this embodiment describes a rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and whale optimization-weighted random forest (WOA-WRF) algorithm:

[0216] Step 1: Feature extraction

[0217] First, characteristic factors that are strongly correlated with the DC breakdown voltage of the rod-plate flame gap are extracted from a large amount of test data. These characteristic factors include but are not limited to:

[0218] Combustion characteristics: including parameters such as flame height, flame width, and flame color, which can reflect the combustion state and intensity of the flame;

[0219] Space temperature: The air temperature around the flame has an important influence on the insulation performance of the air gap;

[0220] Leakage current: The leakage current generated by the flame on the transmission line can reflect the conductivity of the flame on the transmission line;

[0221] Electric field characteristics: including parameters such as electric field intensity and electric field distribution, which can reflect the impact of the electric field on the breakdown voltage;

[0222] When extracting features, advanced feature selection algorithms are used to ensure that the extracted features are both representative and able to reflect the complex changes under wildfire conditions.

[0223] Step 2: Whale Optimization-Weighted Random Forest (WOA-WRF) algorithm

[0224] Next, the Whale Optimization Algorithm (WOA) is used to adaptively optimize the weights of the Weighted Random Forest (WRF) model. As an emerging swarm intelligence optimization algorithm, the Whale Optimization Algorithm has the advantages of fast convergence speed and strong global search capabilities:

[0225] Initialize the whale population: set the initial parameters such as the number of whale populations and the location of whales;

[0226] Whale search process: Based on the principle of the whale optimization algorithm, the whale's position is continuously updated to find the optimal solution. During the search process, the whale will adjust its position based on the distance between the current position and the target position, as well as the interaction force between the whales;

[0227] Determine the termination condition: When the preset number of iterations is reached or other termination conditions are met, the search process stops;

[0228] Output the optimal solution: The optimal solution obtained by the whale optimization algorithm is used as the weight of the weighted random forest model;

[0229] Through the iterative optimization process of the whale optimization algorithm, adaptive adjustment of the weights of the weighted random forest model is achieved, thereby improving the prediction accuracy and adaptability of the model.

[0230] Step 3: Application of prediction model

[0231] Finally, the weighted random forest model adjusted by the whale optimization algorithm is applied to the prediction of the DC breakdown voltage of the rod-plate flame gap. The specific steps are as follows:

[0232] Input the extracted feature factors: Use the feature factors extracted in step 1 as the input of the model;

[0233] Model prediction: The weighted random forest model is used to predict the input characteristic factors and obtain the predicted value of the DC breakdown voltage of the rod-plate flame gap;

[0234] Output prediction results: Output the prediction results to provide reliable technical support for the safe operation of the transmission line.

[0235] In order to verify the effectiveness of the present invention, the following experiments were conducted:

[0236] A section of a power transmission line was selected as the experimental subject, simulating the flame environment of a wildfire. By varying parameters such as flame height, width, and color, as well as adjusting conditions such as space temperature and leakage current, a large amount of experimental data was obtained. The method proposed in this invention then processed and analyzed this experimental data to obtain a predicted value for the DC breakdown voltage of the rod-plate flame gap. Compared with actual results, the prediction results of the present invention are highly accurate and stable.

[0237] Example 5

[0238] In another preferred embodiment, this embodiment, based on Example 4, as another embodiment of the present invention, can further consider the impact of the dynamic characteristics of the flame on the breakdown voltage; for example, dynamic changes such as the jumping and spread of the flame will cause real-time changes in the electric field distribution and the insulation performance of the air gap; in order to capture the impact of these dynamic changes on the breakdown voltage, the extraction of flame dynamic features, such as flame jumping frequency, flame spread speed, etc., can be added to the feature extraction step; then, these dynamic features are input together with other static features into the whale optimization-weighted random forest algorithm for prediction; this can further improve the accuracy and adaptability of the prediction model.

[0239] In the preferred solution, the specific method of Step 3 is to analyze the influence of the input vector characteristics on the prediction results, use the correlation coefficient method to remove the characteristic parameters that are irrelevant or weakly correlated with the breakdown voltage, calculate the Pearson coefficient of each characteristic parameter for the breakdown voltage according to the following formula, and remove redundant features with low correlation to improve the prediction accuracy; the above settings can effectively reduce the complexity of the model and avoid overfitting, thereby improving the model operation efficiency while ensuring the prediction accuracy.

[0240] In the preferred solution, the characteristic factors in Step 4 include combustion characteristics, space temperature, leakage current and electric field characteristics. The combustion characteristics include flame height, flame width and flame color. The above settings can comprehensively reflect multiple dimensions of the equipment operation status. The flame height and width can evaluate the combustion efficiency, and the flame color can reveal whether the combustion is sufficient. Space temperature monitoring helps prevent overheating, and leakage current and electric field characteristics can timely detect electrical safety hazards.

[0241] In the preferred solution, the implementation method for Step 4 is to extract characteristic parameters related to combustion characteristics, spatial temperature, leakage current, electric field characteristics, and breakdown voltage, taking into account factors such as flame bridging, electric field distribution characteristics, and discharge and development channels under flame conditions. The calculation method for these characteristics is also defined. The above settings also require normalization of the extracted characteristic parameters to ensure that each feature has equal weight in subsequent analysis. Simultaneously, a predictive model is established using a machine learning algorithm to predict discharge behavior under flame conditions, providing decision support for safe system operation.

[0242] In the preferred solution, in Step 5, the specific process of the whale optimization algorithm includes: initializing the whale population, the whale search process, and judging the termination condition, and finally outputting the optimal solution as the weight of the weighted random forest model; the above settings gradually approach the global optimum by continuously iterating the whale's position information, wherein the whale search process includes two mechanisms: bubble net attack and random search, to enhance the algorithm's exploration and development capabilities and ensure the accuracy and robustness of the model weights.

[0243] In the preferred solution, the application of the prediction model in Step 6 also includes verification of the accuracy of the prediction model by comparing the predicted values ​​with the actual measured values ​​to evaluate the performance of the prediction model. The above settings can ensure the high reliability of the prediction results and provide solid data support for decision-making. In addition, it is planned to regularly update and optimize the prediction model to adapt to the ever-changing data environment and business needs.

[0244] In the preferred solution, the method further considers the influence of the dynamic characteristics of the flame on the breakdown voltage, and adds the extraction of flame dynamic characteristics, namely the flame jumping frequency and flame spread speed, in the feature extraction Step 1; the above settings enable the model to more accurately predict the changing trend of the breakdown voltage, thereby improving the accuracy and stability of the breakdown voltage prediction.

[0245] In the preferred solution, the dynamic characteristics of the flame are input as additional characteristic factors together with the static characteristic factors into the whale optimization-weighted random forest algorithm model for prediction; the above settings significantly improve the prediction accuracy and stability of the model, making the system more sensitive in identifying changes in flame status, effectively reducing the false alarm rate, and providing strong technical support for the fire warning system.

[0246] In summary, the rod-plate flame gap DC breakdown voltage prediction method and system based on feature extraction and whale optimization-weighted random forest (WOA-WRF) proposed in the present invention demonstrate significant superiority and innovation in the field of electrical engineering power transmission, especially in addressing the problem of inaccurate DC breakdown voltage prediction of the rod-plate flame gap of transmission lines under wildfire conditions. Through the feature extraction step, the present invention carefully screens characteristic factors that are strongly correlated with the breakdown voltage, including combustion characteristics, spatial temperature, leakage current, electric field characteristics, etc., and uses the whale optimization algorithm (WOA) to adaptively optimize the weights of the weighted random forest (WRF) model. This innovative approach not only overcomes the limitations of existing technologies such as low prediction accuracy and poor adaptability, but also significantly improves the prediction accuracy and adaptability of the model.

[0247] During the feature extraction process, the present invention fully considers the impact of the dynamic characteristics of the flame on the breakdown voltage, adding the extraction of flame dynamic characteristics, thereby further improving the accuracy and adaptability of the prediction model. Furthermore, the present invention combines feature extraction technology with advanced machine learning algorithms. This not only considers traditional characteristic factors but also uses multi-physics field coupling simulation to obtain characteristic parameters that are difficult to directly measure experimentally, such as spatial temperature and electric field strength. This allows for a more comprehensive and accurate characterization of transmission line gap breakdown under different operating conditions.

[0248] In view of the data characteristics of gap breakdown under flame conditions, such as multiple features, nonlinearity, and small samples, the present invention creatively proposes a breakdown voltage prediction model based on weighted random forest-whale adaptive optimization. This model combines the ensemble learning advantages of random forests with the global search capability of the whale optimization algorithm, effectively avoiding overfitting of the prediction model and significantly improving the prediction accuracy. By quantifying the influence of different characteristic parameters on the breakdown voltage and eliminating redundant features, the prediction model is further optimized, improving the model's generalization ability and practicality.

[0249] Extensive experimental data and theoretical derivations have fully verified the effectiveness and accuracy of this invention. Compared with existing technologies, the method of this invention not only has higher prediction accuracy but also can adapt to prediction requirements under different operating conditions, providing a more reliable technical guarantee for the safe operation of transmission lines. This innovative achievement not only overcomes the limitations of traditional prediction methods in predicting gap breakdown voltage under flame conditions, but also provides strong technical support for external insulation protection and operation strategy adjustment of transmission lines.

[0250] The implementation of the present invention significantly improves the prediction accuracy and adaptability of the DC breakdown voltage of the transmission line rod-plate flame gap under wildfire conditions, and provides a new solution and technical path for safe operation and reliable prediction in the field of electrical engineering power transmission.

Claims

1. A rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF is characterized by: The method comprises the following steps: Step 1: Build a test platform and conduct a rod-plate gap breakdown test under flame conditions. Obtain the breakdown voltage under typical woodpile sizes, gap distances, and different positive and negative polarity characteristics, and record the corresponding characteristics, including leakage current and environmental factors. Step 2: Establish a gap breakdown simulation. By using a multi-physics field coupling method, establish a gap breakdown simulation corresponding to the experiment in Step 1 to obtain characteristic parameters that are difficult to obtain directly through experiments, including space temperature and electric field strength. Step 3: Eliminate redundant features, quantify the influence of different feature parameters on the breakdown voltage, and eliminate redundant features that are irrelevant or weakly correlated with the breakdown voltage; Step 4: Feature extraction: Extract characteristic factors that are strongly correlated with the DC breakdown voltage of the rod-plate flame gap from the test data, including combustion characteristics, spatial temperature, leakage current, and electric field characteristics. Combustion characteristics include flame height, flame width, and flame color. Step 5: Build a whale optimization-weighted random forest algorithm model, use the whale optimization algorithm to adaptively optimize the weights of the weighted random forest model, and obtain the optimized weighted random forest model; Step 6: Application of the prediction model: input the extracted characteristic factors into the optimized weighted random forest model to obtain the predicted value of the DC breakdown voltage of the rod-plate flame gap.

2. The rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF according to claim 1 is characterized in that: The specific implementation steps in Step 1 include: Step 1.1: Build the test platform, place the woodpile in the middle of the mesh electrode and vertically align it with the rod electrode above, and connect the signal acquisition system to the circuit; Step 1.2: Use a spray bottle to evenly spray industrial alcohol on the woodpile. After igniting the woodpile, turn on the signal acquisition system and video recorder. After the woodpile is burning stably, use the direct lift method to pressurize the gap until it breaks through. Step 1.3: After the flame is extinguished, change the voltage polarity and gap distance, rearrange the woodpile, and repeat the test according to the above steps to obtain the breakdown voltage and leakage current when the flame bridges the rod-plate gap.

3. The rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF according to claim 2 is characterized in that: The specific implementation process of Step 2 is that the main combustible component in the wood pile combustion process is cellulose, and the simplified molecular formula is , then the combustion reaction equation when the pine wood pile undergoes pyrolysis is: （1）； After the woodpile is burned, a large amount of hydrocarbons are generated. The pyrolysis chemical reaction process generates a large amount of heat. Under the action of high heat, the hydrocarbons react with other substances in the air to produce a large number of electrons and positively charged particles. The main ion reactions are as follows: （2）； （3）； The distance between the rod electrode and the woodpile, the radius and length of the rod electrode, and the radius and height of the woodpile are set, and the model geometry simulation is performed to obtain the corresponding electric field distribution and spatial temperature through simulation.

4. The rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF according to claim 3, characterized in that: The specific method of Step 3 is to analyze the influence of the input vector features on the prediction results, use the correlation coefficient method to remove the feature parameters that are irrelevant or weakly correlated with the breakdown voltage, calculate the Pearson coefficient of each feature parameter for the breakdown voltage according to the following formula, and remove redundant features with low correlation to improve the prediction accuracy: （4）； Where, and Represents a characteristic parameter and breakdown voltage No. values; Characteristic parameter The average value of Breakdown voltage The average value of .

5. The rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF according to claim 4 is characterized in that: The implementation method of Step 4 is to consider the flame bridging situation, electric field distribution characteristics, discharge and development channel factors under flame conditions, extract characteristic parameters related to combustion characteristics, spatial temperature, leakage current, electric field characteristics and breakdown voltage, and define the calculation method of related characteristics. The specific implementation steps include: Step 4.1: Select the features on the shortest path to characterize the flame gap breakdown and extract the feature parameters; Step 4.2: The input feature quantities are all taken from one flame oscillation cycle.

6. The rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF according to claim 5, characterized in that: The specific definition of the flame environment characteristic value in Step 4.2 is as follows: Step 4.2.1: Flame root temperature , maximum temperature and average temperature : （5）； （6）； Where, For the root of the flame The temperature of the unit, is the total number of units; Step 4.2.2: Temperature near the rod electrode , maximum temperature and average temperature : （7）； （8）； Where, The rod electrode The temperature of the unit, is the total number of units; Step 4.2.3: Rod-plate gap distance , Maximum flame height Respectively expressed as: （9）； （10）； Where, 、 are the heights of the flame continuous zone and the flame gap zone, respectively. is the coefficient of the flame bridge rod electrode; Step 4.2.4: Leakage current is , respectively take the average value of the leakage current before the gap breakdown and the stable stage , maximum value , standard deviation , leakage current ripple factor It is expressed as follows: （11）； Where, is the effective value of the leakage current; Step 4.2.5: Field strength around the electrode , its maximum field strength is , average field strength : （12）； （13）； Where, The electrode around The temperature of the unit, is the total number of units; Step 4.2.6: Discharge channel field strength , its maximum field strength is , average field strength , energy density : （14）； （15）； （16）； Where, The electrode around The temperature of the unit, is the total number of units; 、 Respectively The energy and volume of a unit.

7. The rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF according to claim 6, characterized in that: In Step 5, the specific process of the whale optimization algorithm includes: initializing the whale population, the whale search process, and judging the termination condition, and finally outputting the optimal solution as the weight of the weighted random forest model. The specific implementation steps of Step 5 include: Step 5.1: Preprocess the sample data, including feature parameter dimensionality reduction, data normalization, and sample sorting to improve the generalization ability and accuracy of the model; Step 5.2: Perform sample extraction, parameter selection and model training, and use Bootstrap resampling to training samples Random sampling with replacement, forming Each subset of samples has different characteristics, and a decision tree is constructed based on the mean square error. Select from the features The binary tree CART is trained with features. In the initial stage, the model prediction result is the mean of the voltage predicted by all decision trees. The out-of-bag data (OOB) that has not been extracted is used as input. The OOB error is calculated to test the model performance. The number of decision trees or the number of feature parameters is changed multiple times until the OOB error no longer changes significantly. Step 5.3: Random Forest uses random sampling and random feature selection to construct a weighted decision tree using the correlation of feature parameters. It then uses the Whale Algorithm (WOA) to adaptively adjust the weight of each decision tree, continuously changing the whale's position, i.e., the decision tree weight. It then approaches the prey through search, encirclement, and predation behaviors, i.e., the optimal solution, until the model's fitness does not improve significantly. The calculation process is as follows: （17）； （18）； Where, For decision tree The weight of is the number of decision trees; For decision tree The sum of the correlation degrees of all features can be expressed as , is the feature in the decision tree The degree of correlation, is the number of features; is the predicted breakdown voltage, For decision tree Predicted breakdown voltage; （19）； Where, is the number of iterations; is the coefficient, expressed as , A random number between 0 and 1; The initial value of is 2 and it is a random number that decreases to 0 during the iteration process; is the current position of the whale; The position of a random whale; is the distance between the whale and its prey, expressed as ; （20）； Where, is the current prey position; Whales travel in spirals: （21）； Where, Used to define the logarithmic spiral shape when the whale updates its position. is a constant; A random number between -1 and 1; The whale's position is continuously updated, and the cycle iterates until the prey is captured. Step 5.4: When establishing a prediction model, in order to improve the prediction accuracy and operation speed of the model, the feature data is processed and the Pearson correlation coefficient method is selected to select the feature parameters. The filtered feature parameters are normalized. The calculation formula is as follows: （22）； Where, is an input parameter; and are the maximum and minimum values ​​of the parameter respectively; is the normalized value; Step 5.5: Select a typical performance evaluation indicator in the random forest regression algorithm, using the root mean square error , coefficient of determination , mean absolute percentage error Overall characterization and evaluation of model performance: （23）； （24）； （25）； Where, For the The actual breakdown voltage of the samples; For the The predicted breakdown voltage of samples; is the total number of samples.

8. The rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF according to claim 7, characterized in that: The application of the prediction model in Step 6 also includes verification of the accuracy of the prediction model by comparing the predicted values ​​with the actual measured values ​​to evaluate the performance of the prediction model.

9. A system for implementing the rod-plate flame gap DC breakdown voltage prediction method based on feature extraction and WOA-WRF as claimed in claim 8, characterized in that: The system comprises: Feature extraction module, used to extract characteristic factors that are strongly correlated with the DC breakdown voltage of the rod-plate flame gap from the test data; The model building module is used to build a whale optimization-weighted random forest algorithm model and use the whale optimization algorithm to adaptively optimize the weights of the weighted random forest model; A prediction module is used to input the extracted characteristic factors into the optimized weighted random forest model to obtain the predicted value of the DC breakdown voltage of the rod-plate flame gap; The verification module is used to verify the accuracy of the prediction model and evaluate the performance of the prediction model by comparing the predicted values ​​with the actual measured values.

Images