Method and device for predicting dissolved gas in oil by considering physical information

Abstract

The invention provides a method and a device for predicting dissolved gas in oil by considering physical information. The method comprises the following steps: carrying out an oil paper insulation discharge test; operating state parameters in the discharging process are collected; collecting an oil sample in the discharge process, and obtaining concentration data of gas dissolved in oil; constructing a neural network prediction model by using the operation state parameters and the concentration data, and performing training; and receiving oil paper insulation discharge real-time monitoring data, and realizing gas production prediction of the dissolved gas in the oil by using the neural network prediction model. Physical information constraint is added to the neural network model, gas production energy, current and the like are used as constraint conditions, the defect that an artificial intelligence data driving model excessively depends on the number and quality of data samples is overcome, gas production of the dissolved gas in the oil can be accurately predicted, and the prediction accuracy is improved. The method has a certain guiding significance for improving the fault diagnosis capability of the dissolved gas in the deep excavation oil and ensuring the safe operation of the power transformer.

Description

Technical Field

[0001] The present invention relates to the technology of predicting and diagnosing dissolved gases in oil, and particularly to a method and device for predicting dissolved gases in oil considering physical information. Background Art

[0002] With the continuous expansion of the scale of the power system and the continuous improvement of the voltage level, as a core device in the power grid, the safe and reliable operation of power transformers is directly related to the stability of the entire power system. The oil-paper insulation system is one of the most commonly used insulation forms in power transformers. In recent years, ester insulation oil has been widely used in power transformers due to its excellent biodegradability, fire resistance and electrical characteristics. However, during long-term operation, the ester insulation oil-paper insulation system may cause insulation aging and faults due to factors such as partial discharge, thus affecting the normal operation of the transformer. Therefore, accurately predicting the generation of dissolved gases in oil (DGA) is crucial for early detection of potential faults and prevention of accidents.

[0003] In addition, although the traditional DGA analysis method based on empirical formulas can reflect the fault state inside the transformer to a certain extent, its diagnostic speed is slow and it is difficult to cope with actual working conditions. In recent years, significant progress has been made in the application of machine learning and deep learning technologies in the field of power equipment fault diagnosis. In particular, neural network models have attracted much attention due to their strong non-linear fitting ability. However, most of the existing neural network models mainly rely on data-driven methods for training, lacking an understanding of physical mechanisms, resulting in problems such as poor generalization ability and insufficient interpretability of the models. Moreover, traditional neural networks have high requirements for data quality and data scale, and often perform poorly in small data scale scenarios.

[0004] Therefore, there is an urgent need to propose a method and device for predicting dissolved gases in oil considering physical information to solve the contradiction between the poor timeliness of current mainstream DGA detection methods, the common problems of poor interpretability, strong dependence on data quality and scale, poor generalization, and easy overfitting of common neural network algorithm models, and the high requirements for timeliness and reliability in modern smart grid fault diagnosis.

[0005] The above information disclosed in this background art section is only used to understand the background art of the inventive concept, and therefore, it may include information that does not constitute prior art. Summary of the Invention

[0006] The object of the present invention is to solve the deficiencies existing in the related technologies in the above-mentioned background technology. Taking synthetic esters as the research object, a method and device for predicting dissolved gases in oil considering physical information are proposed. A neural network model based on physical information is constructed to predict the content of dissolved gases in oil, solving the contradiction between the poor timeliness of the current mainstream DGA detection methods, the poor interpretability, strong dependence on data quality and scale, poor generalization, and easy overfitting of common neural network algorithm models, and the high requirements for timeliness and reliability in modern smart grid fault diagnosis.

[0007] The present invention provides a method and device for predicting dissolved gases in oil considering physical information, including the following steps:

[0008] Step 1: Conduct an oil-paper insulation discharge test;

[0009] Step 2: Collect the operating state parameters during the discharge process;

[0010] Step 3: Collect oil samples during the discharge process and obtain the concentration data of dissolved gases in the oil;

[0011] Step 4: Use the operating state parameters and the concentration data to construct a neural network prediction model and train it;

[0012] Step 5: Receive the real-time monitoring data of oil-paper insulation discharge and use the neural network prediction model to realize the prediction of gas generation of dissolved gases in the oil.

[0013] Preferably, the said Step 1 further includes:

[0014] Establish a set of transformer oil-paper insulation discharge experimental platform to conduct the oil-paper insulation discharge test. The experimental platform includes a discharge test circuit, a test oil tank and a defect model, as well as a signal measurement system.

[0015] Preferably, for the said discharge test circuit, the power supply is a non-corona power frequency test transformer with a rated voltage of 100 kV and a capacity of 20 kVA, and its output voltage can be adjusted by a voltage regulator; the protective resistor uses a water resistor with a resistance value of about 250 MΩ; the voltage divider is a capacitive-resistive voltage divider with a voltage division ratio of 2000:1, which is used to measure high-voltage signals.

[0016] Preferably, for the said test oil tank, the overall box body is made of plexiglass, and is provided with an oil valve, a pressure gauge, and a pressure relief valve. The oil valve is used for oil injection and oil sampling. Electrode grooves are provided at both the upper and lower parts of the oil tank, and the electrodes can be freely replaced.

[0017] Preferably, for the said defect model, a surface discharge defect model is adopted.

[0018] Preferably, the signal measurement system uses a high-frequency current sensor to collect the high-frequency current during the discharge process. Its sensitivity is 10 V / A, the effective detection bandwidth is 45 kHz to 125 MHz, the effective operating temperature is 0 to 65 °C, and a BNC interface is used for signal transmission.

[0019] Preferably, the oil paper is synthetic ester oil paper.

[0020] Preferably, the synthetic ester oil paper selects Midel 7131 synthetic ester, and pretreatment operations such as filtration, drying, and vacuum oil impregnation are performed on the Midel 7131 synthetic ester before the experiment.

[0021] Preferably, the operating state parameters in step 2 include: current data, discharge energy.

[0022] Preferably, obtaining the concentration data of dissolved gases in oil in step 3 specifically includes:

[0023] Taking an oil sample every 15 minutes using the oil valve, and using a gas chromatograph to detect and analyze the components and contents of gases in the insulating oil sample.

[0024] Collecting the contents of 6 gases including hydrogen (H 2 ), carbon monoxide (CO), methane (CH 4 ), ethane (C 2 H 6 ), ethylene (C 2 H 4 ), and acetylene (C 2 H 2 ).

[0025] Step 4 further includes: analyzing the correlation relationship between current, gas production energy, and the production of dissolved gases in oil, and training a neural network considering physical information constraints using a series of data collected during the discharge process to construct a neural network prediction model integrating physical information rules.

[0026] Preferably, constructing a neural network prediction model integrating physical information rules specifically includes:

[0027] Exploring the monotonic relationship between current, gas production energy, and the production of dissolved gases in oil, and this monotonic relationship is used as a physical constraint condition and put into the loss function as a regularization term.

[0028] Specifically, the structural framework of the neural network considering physical information constraints is:

[0029] The number of nodes in the input layer is 8, and the input quantities are current data, discharge energy, and hydrogen (H 2 ), carbon monoxide (CO), methane (CH 4 ), ethane (C2 H 6 ), ethylene (C 2 H 4 ), acetylene (C 2 H 2 ), etc. The number of hidden layer nodes is 12, the number of output layer nodes is 6, and the outputs are the predicted production amounts of six gases, namely hydrogen (H 2 ), carbon monoxide (CO), methane (CH 4 ), ethane (C 2 H 6 ), ethylene (C 2 H 4 ), acetylene (C 2 H 2 ), etc.

[0030] The present invention also discloses a dissolved gas in oil prediction device considering physical information, including: a parameter acquisition functional area, a gas analysis functional area, a model construction functional area, a data prediction functional area, and a visualization interaction functional area.

[0031] Specifically, the parameter acquisition functional area is used to collect multi-dimensional operating state parameters such as voltage and current during the discharge process in real time, as well as the synthetic ester oil sample.

[0032] Specifically, the gas analysis functional area is used to measure the concentration of dissolved gases in the synthetic ester oil sample.

[0033] Specifically, the model construction functional area is used to analyze the complex relationship between current, gas production energy, and the production amount of dissolved gases in oil, and construct a neural network prediction model integrating physical rules.

[0034] Specifically, the data prediction functional area is used to receive and process the real-time monitoring data of synthetic ester oil paper insulation discharge, and realize the dynamic prediction of the production of dissolved gases in oil.

[0035] Specifically, the visualization interaction functional area is used to provide a visualization interface, display the prediction results, analysis reports, and other key data, and support user interaction operations.

[0036] Compared with the related technologies, the beneficial effects of the present invention are as follows: A method and device for predicting dissolved gases in oil considering physical information add physical information constraints to the neural network model, use gas production energy, current, etc. as constraint conditions, avoid the disadvantages of the artificial intelligence data-driven model being overly dependent on the quantity and quality of data samples, and can accurately predict the production of dissolved gases in oil, which has certain guiding significance for improving the fault diagnosis ability of deeply mining dissolved gases in oil and ensuring the safe operation of power transformers.

[0037] The above description is only an overview of the technical solution of the present invention. In order to make the technical means of the present invention clearer and to the extent that those skilled in the art can implement it according to the content of the specification, and in order to make the above and other objects, features, and advantages of the present invention more obvious and understandable, the following will be illustrated by the accompanying drawings and the specific embodiments of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the specific embodiments of the present invention, the accompanying drawings required for the description of the prior art will be briefly introduced below. The accompanying drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. Obviously, the following-described accompanying drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0039] Figure 1 FIG.

[0040] Figure 2 is a schematic flow chart of a method for predicting dissolved gases in oil considering physical information proposed by the present invention;

[0041] Figure 3 FIG.

[0042] Figure 4 is a functional partition diagram of a device for predicting dissolved gases in oil considering physical information designed by the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] The following will refer to Figures 1 to 4 to describe the specific embodiments of the present invention in more detail. Although the specific embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art. It should be understood that these embodiments are only for illustrating the present invention and not for limiting the scope of the present invention.

[0044] In addition, it should be understood that the present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0045] For better understanding of the embodiments of the present invention, a specific embodiment will be further explained below in conjunction with the accompanying drawings, and the accompanying drawings do not limit the embodiments of the present invention.

[0046] In one embodiment, as Figure 1 shown, a method for predicting dissolved gases in oil considering physical information specifically includes:

[0047] Step 1: Conduct a synthetic ester oil-paper insulation discharge test;

[0048] Step 2: Collect operating state parameters such as voltage and current during the discharge process;

[0049] Step 3: Collect synthetic ester oil samples during the discharge process and obtain concentration data of dissolved gases in the oil;

[0050] Step 4: Analyze the correlation between current, gas production energy and the production of dissolved gases in the oil, and use a series of data collected during the discharge process to construct and train a neural network model considering physical information constraints;

[0051] Step 5: Receive real-time monitoring data of synthetic ester oil-paper insulation discharge, realize the prediction of dissolved gas production in the oil, and compare with other models to highlight the advantages of the neural network model considering physical information constraints.

[0052] Further, in Step 1, a synthetic ester oil-paper insulation discharge test is conducted.

[0053] Specifically, a set of transformer oil-paper insulation discharge experimental platforms is established to carry out synthetic ester oil-paper insulation discharge tests. The experimental platform consists of a discharge test circuit, a test oil tank and a defect model, and a signal measurement system.

[0054] In another embodiment, the discharge test circuit is as Figure 4 shown. The power supply is a non-corona power frequency test transformer with a rated voltage of 100 kV and a capacity of 20 kVA, and its output voltage can be adjusted by a voltage regulator; the protection resistor uses a water resistor with a resistance value of about 250 MΩ; the voltage divider is a capacitive voltage divider, specifically a capacitive-resistive voltage divider with a voltage division ratio of 2000:1, which is used to measure high-voltage signals.

[0055] In another embodiment, the test oil tank has an overall body made of plexiglass and is provided with an oil valve, a pressure gauge, and a pressure relief valve. The oil valve is used for oil injection and oil sampling, and electrode grooves are provided at both the upper and lower parts of the oil tank, and the electrodes can be freely replaced.

[0056] In another embodiment, the defect model adopts a surface discharge defect model.

[0057] In another embodiment, the signal measurement system uses a high-frequency current sensor to collect the high-frequency current during the discharge process. Its sensitivity is 10 V / A, the effective detection bandwidth is 45 kHz to 125 MHz, the effective operating temperature is 0 to 65 °C, and a BNC interface is used for signal transmission.

[0058] In another embodiment, the synthetic ester insulating paper selects Midel 7131 synthetic ester, and pretreatment operations such as filtration, drying, and vacuum oil impregnation are performed on the Midel 7131 synthetic ester before the experiment.

[0059] Furthermore, in step 2, operating state parameters such as voltage and current during the discharge process are collected.

[0060] In another embodiment, a capacitive voltage divider with a voltage division ratio of 2000:1 is used to collect the voltage signal during the discharge process.

[0061] In another embodiment, a high-frequency current sensor is used to collect the current signal during the discharge process.

[0062] Furthermore, in step 3, a synthetic ester oil sample is collected during the discharge process, and the concentration data of the dissolved gases in the oil is obtained.

[0063] Specifically, the oil valve is used to take an oil sample every 15 minutes, and a gas chromatograph is used to detect and analyze the components and contents of the gases in the insulating oil sample.

[0064] In another embodiment, the contents of 6 gases, namely hydrogen (H 2 ), carbon monoxide (CO), methane (CH 4 ), ethane (C 2 H 6 ), ethylene (C 2 H 4 ), and acetylene (C 2 H 2 ), are collected respectively.

[0065] Furthermore, in step 4, the correlation relationship between current, gas generation energy, and the production of dissolved gases in the oil is analyzed, and a neural network model considering physical information constraints is constructed and trained using the series of data collected during the discharge process.

[0066] Specifically, the standard enthalpy of formation of H 2 is 128.5 kJ / mol, the standard enthalpy of formation of CO is 110.5 kJ / mol, the standard enthalpy of formation of CH 4 is 77.7 kJ / mol, the standard enthalpy of formation of C 2 H 6 is 93.5 kJ / mol, the standard enthalpy of formation of C2 H 4 The standard enthalpy of formation of 2 H 2 is 104.1 kJ / mol, and the standard enthalpy of formation of the

[0067] The calculation formula for the gas production energy E is as follows:

[0068]

[0069] wherein, represents the gas content concentration of hydrogen, and the same applies to the other gases.

[0070] In the embodiment, the current values (mA), gas production energy (kJ / L), and the content concentrations of solvent gases in oil (μL / L) at three discharge moments were collected as shown in the following table:

[0071]

[0072] Specifically, the monotonic relationship among the current, gas production energy, and the production of dissolved gases in oil was explored. This monotonic relationship is the physical constraint condition, which is used as a regular term in the loss function.

[0073] The monotonic relationship is as follows: during the discharge process, as the current value increases and the gas production energy increases, the production of dissolved gases in oil also increases, showing a positive correlation.

[0074] In another embodiment, as shown in Figure 2 , the structural framework of the neural network considering physical information constraints is as follows:

[0075] The number of input layer nodes is 8, and the input quantities are current data, discharge energy, and the contents of 6 gases such as hydrogen (H 2 ), carbon monoxide (CO), methane (CH 4 ), ethane (C 2 H 6 ), ethylene (C 2 H 4 ), and acetylene (C 2 H 2 ). The number of hidden layer nodes is 12, and the number of output layer nodes is 6. The output quantities are the predicted production amounts of 6 gases such as hydrogen (H 2 ), carbon monoxide (CO), methane (CH 4 ), ethane (C 2 H 6 ), ethylene (C 2 H 4 ), and acetylene (C 2 H 2 ).

[0076] In another embodiment, the training process of the neural network considering physical information constraints is as follows: The input layer inputs data, the hidden layer receives the data transmitted by the input layer, performs a non-linear transformation on the data using the activation function tanh, and transmits it to the output layer. The output layer outputs the prediction result. In this network structure, the input layer is a fully connected layer that receives the input data and directly passes it to the hidden layer; the hidden layer is also composed of fully connected layers, calculates through the weight connections between nodes, completes the non-linear transformation and extracts features in combination with the tanh activation function; the output layer is also a fully connected layer that maps the features of the hidden layer to the prediction result. The entire network does not involve pooling or recursive structures, but is based on the layer-by-layer transmission of fully connected layers, calculates the output through forward propagation, optimizes by combining the prediction error and the physical constraint residual through the loss function, and uses the gradient descent method to iteratively update the weights to minimize the objective function.

[0077] Furthermore, the tanh activation function is used for non-linear conversion between layers; when calculating the error, the data error is calculated by comparing the prediction result with the true value, and the physical quantity predicted by the network is substituted into the corresponding physical constraint to calculate the residual. The sum of the two is the error obtained by the loss function; the error is used for backpropagation for multiple iterations to update the weights between the nodes on the network. The optimization algorithm uses the gradient descent method and stops iterating when the value of the loss function drops to a threshold of 10e -4 as follows. It should be noted that during the model training process, the training data used are all obtained through real experiments and form an expert database. Before training the model, data collection and recording are completed through multiple experiments. Specifically, the experiment includes steps 1 to 3. In each experiment, the concentrations, currents, and discharge energy data of 6 gases are obtained through oil sample collection and used as the model input to predict the future concentration values of 6 gases, that is, the predicted values; while the actual concentration values of the 6 gases obtained by collecting oil samples in the experiment are the true values. Through multiple repeated experiments, an expert database containing rich experimental data is constructed for training the neural network model. After training, the model can accurately predict the input data of newly collected oil samples.

[0078] The data error between the predicted value and the true value is calculated by the formula:

[0079]

[0080] In the formula, represents the true value; represents the model predicted value; represents the time step index, and the value range is [1, N - 1]; Denote the output prediction value index, whose value range is [1, 6]; N represents the total number of time steps, that is, the total number of steps included in the time series data. In this embodiment, N refers to the number of times of taking oil samples, that is, N oil samples are taken at N different times.

[0081] The residual obtained by physical constraint calculation The calculation formula is:

[0082]

[0083] In the formula, Denote the time step index, whose value range is [1, N - 1]; and Denote two consecutive time points; Denote the output prediction value index, whose value range is [1, 6]; is a positive number approximately equal to 0 to prevent the denominator from being zero; represents current; represents gas production energy; represents the predicted value of dissolved gas in oil.

[0084] If the monotonic relationship is satisfied among current, gas production energy and the production of dissolved gas in oil, the loss function L only includes data error; if the monotonic relationship is not satisfied, the loss function L is the sum of data error and the residual obtained by physical constraint calculation to punish the violation of monotonicity. That is:

[0085]

[0086] In the formula, is a weight parameter used to adjust the influence of physical constraint residual and data error. In the embodiment of the present invention, is taken.

[0087] Furthermore, in step 5, receive the real-time monitoring data of the synthetic ester oil-paper insulation discharge, realize the gas production prediction of dissolved gas in oil, and compare with other models to highlight the advantages of the neural network model considering physical information constraints.

[0088] In another embodiment, use the same data to train a general neural network model without considering physical information constraints, and evaluate it through the coefficient of determination index to highlight the advantages of the neural network model considering physical information constraints.

[0089] Coefficient of determination The calculation formula is:

[0090]

[0091] In the formula, represents the true value, represents the average value of the true values, and represents the predicted values of the model.

[0092] In another embodiment, 450 groups of data are used to separately train a neural network model considering physical information constraints and a general neural network model without considering physical information constraints, 150 groups of data are used to test the two models, and the coefficient of determination is calculated.

[0093] In this embodiment, the coefficient of determination of the neural network model considering physical information constraints is , and the coefficient of determination of the general neural network model without considering physical information constraints is . By comparison, it can be seen that the neural network model considering physical information constraints designed by the present invention can not only accurately predict the dissolved gas data in oil, but also, due to making full use of the existing physical prior knowledge for constraint, has stronger robustness and generalization ability when dealing with small-scale data, and the interpretability of the model inference process is also improved.

[0094] In addition, the present invention also proposes an oil dissolved gas prediction device considering physical information, as Figure 3 shown, the device includes: a parameter acquisition functional area Div1, a gas analysis functional area Div2, a model construction functional area Div3, a data prediction functional area Div4, and a visualization interaction functional area Div5.

[0095] Specifically, the parameter acquisition functional area Div1 is used to collect multi-dimensional operating state parameters such as voltage and current during the discharge process in real time and synthetic ester oil samples.

[0096] Specifically, the gas analysis functional area Div2 is used to measure the concentration of dissolved gases in the synthetic ester oil sample.

[0097] Specifically, the model construction functional area Div3 is used to analyze the complex relationship between current, gas production energy and the production of dissolved gases in oil, and construct and train a neural network model considering physical information constraints.

[0098] Specifically, the data prediction functional area Div4 is used to receive and process the real-time monitoring data of the synthetic ester oil-paper insulation discharge, and realize the dynamic prediction of the production of dissolved gases in oil.

[0099] Specifically, the visualization interaction functional area Div5 is used to provide a visualization interface, display prediction results, analysis reports and other key data, and support user interaction operations.

[0100] In summary, the present invention discloses a method and device for predicting dissolved gases in oil considering physical information, including step 1 of conducting a synthetic ester oil-paper insulation discharge test; step 2 of collecting operating state parameters such as voltage and current during the discharge process; step 3 of collecting synthetic ester oil samples during the discharge process and obtaining concentration data of dissolved gases in the oil; step 4 of analyzing the correlation between current, gas production energy, and the production of dissolved gases in the oil, and constructing and training a neural network model considering physical information constraints using a series of data collected during the discharge process; step 5 of receiving real-time monitoring data of synthetic ester oil-paper insulation discharge, realizing the prediction of dissolved gas production in the oil, and comparing with other models to highlight the advantages of the neural network model considering physical information constraints. The present invention adds physical information constraints to the neural network model, using gas production energy, current, etc. as constraint conditions, avoiding the disadvantages of artificial intelligence data-driven models being overly dependent on the quantity and quality of data samples, and being able to accurately predict the production of dissolved gases in the oil, which has certain guiding significance for improving the fault diagnosis ability of deeply mining dissolved gases in the oil and ensuring the safe operation of power transformers.

[0101] The above embodiments are only for illustrating the technical concept and characteristics of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly, and should not be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting dissolved gas in oil considering physical information, characterized in that: The following steps are involved: Step 1: Conduct oil-paper insulation discharge test; Step 2: Collect operating state parameters during the discharge process; Step 3: Collect oil samples during the discharge process and obtain the concentration data of dissolved gas in the oil; Step 4: construct a neural network prediction model using the operating state parameters and the concentration data and perform training; Step 5: Receive the real-time monitoring data of oil-paper insulation discharge, and use the neural network prediction model to predict the gas production of dissolved gas in oil.

2. The method according to claim 1, characterized in that Preferably, the step 1 further comprises: A transformer oil-paper insulation discharge experimental platform is established to carry out oil-paper insulation discharge tests. The experimental platform includes a discharge test circuit, a test oil tank, a defect model and a signal measurement system.

3. The method according to claim 2, characterized in that The power supply of the discharge test circuit is a non-corona power frequency test transformer, and its output voltage can be adjusted by a voltage regulator.

4. The method according to claim 2, characterized in that: The test oil tank has an entire body made of organic glass and is provided with an oil valve, a pressure gauge, and a pressure release valve.

5. The method according to claim 2, characterized in that: The defect model adopts a surface discharge defect model.

6. The method according to claim 2, characterized in that The signal measurement system uses a high-frequency current sensor to collect high-frequency current during the discharge process.

7. The method according to claim 1, characterized in that The oil paper is synthetic ester oil paper.

8. The method according to claim 7, characterized in that The synthetic ester oil paper is Midel 7131 synthetic ester.

9. The method according to claim 1, characterized in that: The operating state parameters specifically include: current data and discharge energy.

10. A device for predicting dissolved gas in oil taking physical information into consideration, characterized in that: include: Parameter collection function area, gas analysis function area, model building function area, data prediction function area, and visualization interaction function area; The parameter acquisition functional area is used to collect multi-dimensional operating state parameters such as voltage and current and synthetic ester oil samples during the discharge process in real time; The gas analysis functional area is used to measure the concentration of dissolved gas in the synthetic ester oil sample; The model constructs a functional area for analyzing the complex relationship between current, gas production energy and the production of dissolved gas in oil, and constructs a neural network prediction model integrating physical rules; The data prediction functional area is used to receive and process the real-time monitoring data of the oil-paper insulation discharge test to achieve dynamic prediction of the gas production of dissolved gas in the oil; The visualization interactive function area is used to provide a visualization interface, display prediction results, analysis reports and other key data, and support user interactive operations.

Images