A method and system for predicting moisture content of transformer oil

Abstract

The present invention discloses a method and system for predicting the moisture content of transformer oil, comprising: obtaining a first ultrasonic soundprint feature of the transformer oil to be tested in real time; inputting the first ultrasonic soundprint feature into a moisture content prediction model, and using the first moisture content output by the model as a moisture content prediction result of the transformer oil to be tested; wherein the first ultrasonic soundprint feature includes a first sound velocity parameter, a first maximum amplitude, a first absolute average value, a first root mean square value, and a first waveform index; the moisture content prediction model is obtained by training and verifying an RBF neural network model based on a first sample data set, and the first sample data set includes several groups of second ultrasonic soundprint features and a second moisture content corresponding to each group of second ultrasonic soundprint features. The present invention uses the moisture content prediction model to analyze the moisture content prediction results of the transformer oil to be tested in real time from two aspects: propagation characteristics and time domain characteristics, to ensure the real-time and accuracy of the moisture content prediction.

Description

Technical Field

[0001] The present invention relates to the field of transformer oil quality detection, and in particular to a method and system for predicting the moisture content of transformer oil. Background Art

[0002] Power transformers, responsible for regulating voltage levels, transmitting energy, and controlling power flows, are core electrical equipment in power systems. Their stable, safe, efficient, and long-term operation is directly related to the stability and safety of the power system. A transformer failure can have serious consequences, leading to widespread power outages, significant economic losses, and even serious safety incidents. Therefore, inspecting the operating status of transformer insulation and promptly detecting and resolving potential faults is a crucial measure for ensuring power system safety and efficiency. As one of the most important insulating media within a transformer, transformer oil is responsible for insulation, heat dissipation, and arc suppression. Its quality not only affects the transformer's electrical characteristics but also significantly impacts its service life. Moisture content is one of the key indicators for evaluating transformer oil quality. Excessive moisture content in transformer oil can lead to numerous insulation system problems, such as reduced breakdown voltage, increased dielectric loss, and accelerated chemical reactions of organic matter, potentially leading to insulation breakdown and equipment burnout. Therefore, monitoring transformer oil moisture content during operation is crucial for maintaining stable operation.

[0003] Currently, the power industry typically removes insulating oil from operating transformers and regularly measures its moisture content in external laboratories using Karl Fischer titration. While this method offers high measurement accuracy, the measurement process is cumbersome, making it difficult to detect potential transformer faults in a timely manner. Furthermore, the oil is susceptible to external contamination during transportation to the external laboratory. Summary of the Invention

[0004] The present invention provides a method and system for predicting the moisture content of transformer oil, which further improves the prediction accuracy of the moisture content of transformer oil while ensuring the real-time prediction of the moisture content of transformer oil.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a method for predicting the moisture content of transformer oil, comprising:

[0006] Acquire the first ultrasonic soundprint feature of the transformer oil to be tested in real time;

[0007] Inputting the first ultrasonic voiceprint feature into a preset moisture content prediction model, and using the first moisture content output by the moisture content prediction model as a moisture content prediction result of the transformer oil to be tested, so as to complete the moisture content prediction of the transformer oil to be tested;

[0008] Among them, the first ultrasonic voiceprint feature includes a first sound speed parameter, a first maximum amplitude, a first absolute average value, a first root mean square value and a first waveform index. The moisture content prediction model is obtained by training and verifying the RBF neural network model based on the first sample data set. The first sample data set includes several groups of second ultrasonic voiceprint features and the second moisture content corresponding to each group of the second ultrasonic voiceprint features.

[0009] Implementation of the present invention utilizes a mapping relationship between ultrasonic soundprint features and moisture content established by a moisture content prediction model to predict and output the moisture content of the transformer oil under test in real time. This ensures the real-time nature of the transformer oil moisture content prediction, thereby helping users promptly identify potential transformer faults without interference from the external environment. Furthermore, by using the first sound velocity parameter, the first maximum amplitude, the first absolute average value, the first root mean square value, and the first waveform index as the basis for predicting the moisture content of the transformer oil, the moisture content prediction model analyzes the moisture content prediction results of the transformer oil under test from both propagation characteristics and time domain characteristics, thereby improving the generalization capability and prediction accuracy of the moisture content prediction model.

[0010] As a preferred solution, the moisture content prediction model is obtained as follows:

[0011] Performing real-time testing on the transformer oil to be tested in different working conditions to obtain several groups of second ultrasonic soundprint features corresponding to the transformer oil to be tested, and using a coulometric trace water meter to perform real-time testing on the transformer oil to be tested in different working conditions to obtain the second moisture content corresponding to each group of the second ultrasonic soundprint features, and then using all the second ultrasonic soundprint features and the second moisture content corresponding to each group of the second ultrasonic soundprint features as the first sample data set;

[0012] Based on the RBF neural network, the RBF neural network model is constructed, and the RBF neural network model is trained and verified using the first sample data set to obtain the moisture content prediction model;

[0013] Among them, the second ultrasonic voiceprint feature includes a second sound speed parameter, a second maximum amplitude, a second absolute average value, a second root mean square value and a second waveform index.

[0014] By implementing a preferred embodiment of the present invention, by acquiring multiple sets of second ultrasonic soundprint features corresponding to transformer oil under different operating conditions in real time and using a coulometric trace moisture meter to detect the second moisture content corresponding to each set of second ultrasonic soundprint features, these features are used as a first sample dataset to train and verify the RBF neural network model, further improving the prediction accuracy of the moisture content prediction model. Furthermore, because RBF neural networks can approximate arbitrary nonlinear functions with arbitrary precision and possess global approximation capabilities, the RBF neural network model constructed based on the RBF neural network exhibits rapid training convergence and strong generalization capabilities, thereby improving the training speed and prediction performance of the moisture content prediction model.

[0015] As a preferred solution, the RBF neural network model is constructed based on the RBF neural network, and the RBF neural network model is trained and verified using the first sample data set to obtain the moisture content prediction model, specifically:

[0016] Dividing all the second ultrasonic voiceprint features in the first sample data set and the second moisture content corresponding to each group of the second ultrasonic voiceprint features to obtain a training data set and a test data set corresponding to the first sample data set;

[0017] Based on the RBF neural network, the RBF neural network model is constructed, and the RBF neural network model is trained using the training data set to form an initial mapping relationship between ultrasonic voiceprint features and moisture content, thereby obtaining an initial detection model;

[0018] Inputting each group of the second ultrasonic voiceprint features in the test data set into the initial detection model to obtain a moisture content prediction result for each second ultrasonic voiceprint feature, and calculating the moisture content prediction accuracy of the initial detection model based on the moisture content prediction results of all the second ultrasonic voiceprint features and the second moisture contents corresponding to all the second ultrasonic voiceprint features, and determining whether the moisture content prediction accuracy meets a preset range;

[0019] If so, the training and verification of the RBF neural network model is completed, and the current initial detection model is used as the moisture content prediction model;

[0020] If not, obtain multiple groups of third ultrasonic voiceprint features corresponding to the transformer oil to be tested under different working conditions, and the third moisture content corresponding to each of the third ultrasonic voiceprint features, and use all the third ultrasonic voiceprint features and the third moisture content corresponding to each of the third ultrasonic voiceprint features as the second sample data set, and iteratively train and verify the current initial detection model based on the current second sample data set to obtain the updated initial detection model, until the moisture content prediction accuracy of the current initial detection model meets the preset range, and then use the current initial detection model as the moisture content prediction model.

[0021] According to a preferred embodiment of the present invention, a training data set is used to train an RBF neural network model, and the output results of the initial detection model obtained through training are compared with the actual measurement data in the test data set to verify the moisture content prediction accuracy of the initial detection model. Meanwhile, if the moisture content prediction accuracy does not meet the preset range, a new sample data set is obtained and the RBF neural network model is trained multiple times using the new sample data set until the moisture content prediction accuracy meets the preset range, so as to optimize and adjust the model parameters so that the moisture content prediction model finally obtained can be applied to the moisture content prediction of various transformer oils.

[0022] As a preferred solution, the real-time acquisition of the first ultrasonic voiceprint feature of the transformer oil to be tested is specifically as follows:

[0023] Using a detection platform, the ultrasonic signal propagating in the transformer oil to be tested is collected and converted in real time to obtain an electrical signal corresponding to the ultrasonic signal;

[0024] On the oscilloscope of the detection platform, waveform information corresponding to the electrical signal is displayed in real time, and according to a preset algorithm, the first sound velocity parameter, the first maximum amplitude, the first absolute average value, the first root mean square value, and the first waveform index corresponding to the transformer oil to be tested are calculated in combination with the waveform information;

[0025] The detection platform includes a signal generator, an ultrasonic power amplifier, the oscilloscope, a piezoelectric ultrasonic transmitting transducer, a piezoelectric ultrasonic receiving transducer and a fuel tank.

[0026] By implementing the preferred solution of the embodiment of the present invention, the ultrasonic signal collected in real time is converted into an electrical signal, and the waveform information corresponding to the electrical signal is displayed in real time on an oscilloscope. Thus, by analyzing and calculating the waveform information, the propagation characteristics and time domain characteristics of the ultrasonic signal can be analyzed to obtain parameters for evaluating the performance of the transformer oil to be tested, thereby reducing the processing difficulty of directly analyzing the ultrasonic signal and improving the overall prediction efficiency.

[0027] As a preferred solution, the RBF neural network model includes: an input layer, a hidden layer and an output layer;

[0028] The input layer includes five input neurons for sequentially receiving the first sound velocity parameter, the first maximum amplitude, the first absolute average value, the first root mean square value, and the first waveform index of the transformer oil to be tested;

[0029] The hidden layer includes a plurality of hidden neurons, and the transformation function of each hidden neuron is a radial basis function;

[0030] The output layer includes an output neuron, which is used to output the moisture content prediction result of the transformer oil to be tested.

[0031] By implementing the preferred solution of the embodiment of the present invention, the simple architecture of the RBF neural network model can accelerate the speed of training and verifying the RBF neural network model, and accelerate the prediction speed of the moisture content prediction model obtained by training and verification, so as to further ensure the real-time prediction of the moisture content of the transformer oil.

[0032] In order to solve the same technical problem, an embodiment of the present invention further provides a system for predicting moisture content of transformer oil, comprising:

[0033] A first data acquisition module is configured to acquire, in real time, a first ultrasonic voiceprint feature of the transformer oil to be tested; wherein the first ultrasonic voiceprint feature includes a first sound velocity parameter, a first maximum amplitude, a first absolute average value, a first root mean square value, and a first waveform index;

[0034] A prediction module is used to input the first ultrasonic voiceprint feature into a preset moisture content prediction model, and use the first moisture content output by the moisture content prediction model as the moisture content prediction result of the transformer oil to be tested, so as to complete the moisture content prediction of the transformer oil to be tested; wherein, the moisture content prediction model is obtained by training and verifying the RBF neural network model based on a first sample data set, and the first sample data set includes several groups of second ultrasonic voiceprint features and the second moisture content corresponding to each group of the second ultrasonic voiceprint features.

[0035] As a preferred embodiment, the transformer oil moisture content prediction system further includes:

[0036] a second data acquisition module, configured to perform real-time detection on the transformer oil to be tested in different working conditions to obtain several groups of second ultrasonic soundprint features corresponding to the transformer oil to be tested, and use a coulometric trace water meter to perform real-time detection on the transformer oil to be tested in different working conditions to obtain the second moisture content corresponding to each group of the second ultrasonic soundprint features, and then use all the second ultrasonic soundprint features and the second moisture content corresponding to each group of the second ultrasonic soundprint features as the first sample data set; wherein the second ultrasonic soundprint features include a second sound velocity parameter, a second maximum amplitude, a second absolute average value, a second root mean square value, and a second waveform index;

[0037] The model building module is used to build the RBF neural network model based on the RBF neural network, and use the first sample data set to train and verify the RBF neural network model to obtain the moisture content prediction model.

[0038] As a preferred solution, the model building module specifically includes:

[0039] a data partitioning unit, configured to partition all the second ultrasonic voiceprint features in the first sample data set, and the second moisture content corresponding to each group of the second ultrasonic voiceprint features, to obtain a training data set and a test data set corresponding to the first sample data set;

[0040] A model training unit is used to construct the RBF neural network model based on the RBF neural network, and train the RBF neural network model using the training data set to form an initial mapping relationship between ultrasonic voiceprint features and moisture content, thereby obtaining an initial detection model;

[0041] A model verification unit is used to input each group of the second ultrasonic voiceprint features in the test data set into the initial detection model respectively to obtain the moisture content prediction results of each second ultrasonic voiceprint feature, and calculate the moisture content prediction accuracy of the initial detection model according to the moisture content prediction results of all the second ultrasonic voiceprint features and the second moisture content corresponding to all the second ultrasonic voiceprint features, and judge whether the moisture content prediction accuracy meets the preset range; if the moisture content prediction accuracy meets the preset range, the training verification of the RBF neural network model is completed, and the current initial detection model is used as the moisture content prediction model; if If the moisture content prediction accuracy does not meet the preset range, multiple groups of third ultrasonic voiceprint features corresponding to the transformer oil to be tested under different working conditions and the third moisture content corresponding to each of the third ultrasonic voiceprint features are obtained, and all the third ultrasonic voiceprint features and the third moisture content corresponding to each of the third ultrasonic voiceprint features are used as the second sample data set, and the current initial detection model is iteratively trained and verified based on the current second sample data set to obtain the updated initial detection model, until the moisture content prediction accuracy of the current initial detection model meets the preset range, and the current initial detection model is used as the moisture content prediction model.

[0042] As a preferred solution, the data acquisition module specifically includes:

[0043] An acquisition and conversion unit is used to acquire and convert the ultrasonic signal propagating in the transformer oil to be tested in real time using a detection platform to obtain an electrical signal corresponding to the ultrasonic signal; wherein the detection platform includes a signal generator, an ultrasonic power amplifier, an oscilloscope, a piezoelectric ultrasonic transmitting transducer, a piezoelectric ultrasonic receiving transducer, and an oil tank;

[0044] a computing unit configured to display waveform information corresponding to the electrical signal on the oscilloscope of the detection platform in real time, and to calculate, in accordance with a preset algorithm and in combination with the waveform information, the first sound velocity parameter, the first maximum amplitude, the first absolute average value, the first root mean square value, and the first waveform index corresponding to the transformer oil to be tested.

[0045] As a preferred solution, the RBF neural network model includes: an input layer, a hidden layer and an output layer;

[0046] The input layer includes five input neurons for sequentially receiving the first sound velocity parameter, the first maximum amplitude, the first absolute average value, the first root mean square value, and the first waveform index of the transformer oil to be tested;

[0047] The hidden layer includes a plurality of hidden neurons, and the transformation function of each hidden neuron is a radial basis function;

[0048] The output layer includes an output neuron, which is used to output the moisture content prediction result of the transformer oil to be tested. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 : A schematic flow chart of a method for predicting moisture content in transformer oil provided in Example 1 of the present invention;

[0050] Figure 2 : A schematic diagram of the structure of the detection platform provided in Example 1 of the present invention;

[0051] Figure 3 : A structural diagram of a transformer oil moisture content prediction system provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0053] Example 1:

[0054] Please refer to Figure 1 , a method for predicting the moisture content of transformer oil provided by an embodiment of the present invention, the method includes steps S1 to S2, each step is specifically as follows:

[0055] Step S1, obtaining a first ultrasonic soundprint feature of the transformer oil to be tested in real time; wherein the first ultrasonic soundprint feature includes a first sound velocity parameter, a first maximum amplitude, a first absolute average value, a first root mean square value and a first waveform index.

[0056] As a preferred solution, step S1 includes steps S11 to S12, and the details of each step are as follows:

[0057] In step S11 , the ultrasonic signal propagating in the transformer oil to be tested is collected and converted in real time using the detection platform to obtain an electrical signal corresponding to the ultrasonic signal.

[0058] Step S12: Displaying the waveform information corresponding to the electrical signal in real time on the oscilloscope of the detection platform.

[0059] In this embodiment, please refer to Figure 2The detection platform specifically includes a signal generator 1, an ultrasonic power amplifier 2, an oscilloscope 3, a piezoelectric ultrasonic transmitting transducer 4, a piezoelectric ultrasonic receiving transducer 5, a fuel tank 6, and a computer 7. Before activating the detection platform, certain preparatory measures must be taken, including steps ① to ②. It is worth noting that the order of steps ① to ② is not necessarily limited to this. The order can be reversed, or individual steps can be performed simultaneously. This is not specifically limited here and depends on the specific application environment. The specific steps are as follows:

[0060] Step ①, invert the transformer oil to be tested in the oil tank 6, then immerse the piezoelectric ultrasonic transmitting transducer 4 and the piezoelectric ultrasonic receiving transducer 5 in the medium of the transformer oil to be tested, and keep the distance L between the piezoelectric ultrasonic transmitting transducer 4 and the piezoelectric ultrasonic receiving transducer 5 at 8 cm, and both are on the same horizontal plane.

[0061] Step ②, set the signal generator 1 to: a sine wave waveform, a frequency of 1 MHz, and a peak-to-peak value of 4 V, set the amplification factor of the first channel of the power amplifier 2 (i.e., the port connected to the piezoelectric ultrasonic transmitting transducer 4) to 40 dB, and set the amplification factor of the second channel of the power amplifier 2 (i.e., the port connected to the piezoelectric ultrasonic receiving transducer 5) to 20 dB.

[0062] It should be noted that the detection principle of the detection platform is specifically as follows: an excitation signal is emitted by a signal generator 1, and the excitation signal is amplified by an ultrasonic power amplifier 2, so that the piezoelectric ceramic piece in the piezoelectric ultrasonic transmitting transducer 4 resonates with a vibration center frequency of 1 MHz, thereby driving the piezoelectric ultrasonic transmitting transducer 4 to work normally and emit an ultrasonic signal, and the ultrasonic signal is transmitted in the transformer oil to be detected stored in the oil tank 6; the piezoelectric ultrasonic receiving transducer 5 uses a transmission method to collect the ultrasonic signal and convert it into a corresponding electrical signal, and then the electrical signal is amplified by the ultrasonic power amplifier 2; the waveform information corresponding to the electrical signal is displayed in real time by an oscilloscope 3 with a bandwidth of 70 MHz and a signal sampling rate of 8 GSa / s, and the electrical signal is input into a computer 7 based on the USB Host & Device and HDMI interfaces of the oscilloscope 3 for subsequent signal analysis, processing and storage.

[0063] Signal generator 1 has a high frequency stability of ±1 ppm, a phase noise as low as -105 dBc / Hz, a maximum frequency of 50 MHz, and a sampling rate of 250 MSa / s. Oscilloscope 3 has a memory depth of 200 Mpts, and fuel tank 6 has dimensions of 30 cm × 20 cm × 15 cm. Furthermore, waveform information is obtained by averaging multiple measurements of the ultrasonic signal waveform, with the number of waveform sampling points set to n = 1000.

[0064] Step S13: According to the preset algorithm, combined with the waveform information, the first sound velocity parameter c and the first maximum amplitude u corresponding to the transformer oil to be tested are calculated respectively. p , the first absolute mean The first root mean square value u ms and the first waveform index W, whose calculation formulas are respectively shown in formulas (1)(2)(3)(4)(5).

[0065]

[0066] u p =max|u(i)| (2)

[0067]

[0068]

[0069]

[0070] Wherein, L is the distance between the piezoelectric ultrasonic transmitting transducer 4 and the piezoelectric ultrasonic receiving transducer 5, t0 is the time from the first amplitude of the piezoelectric ultrasonic transmitting transducer 4 to the first amplitude signal of the piezoelectric ultrasonic receiving transducer 5, u(i) is the voltage value of the i-th sampling value (i=1, 2..., n), and n is the number of sampling points.

[0071] Step S2, inputting the first ultrasonic voiceprint feature into a preset moisture content prediction model, and using the first moisture content output by the moisture content prediction model as the moisture content prediction result of the transformer oil to be tested, so as to complete the moisture content prediction of the transformer oil to be tested; wherein, the moisture content prediction model is obtained by training and verifying the RBF neural network model based on the first sample data set, and the first sample data set includes several groups of second ultrasonic voiceprint features and the second moisture content corresponding to each group of second ultrasonic voiceprint features.

[0072] As a preferred solution, the structure of the RBF (Radical Basis Function) neural network model mentioned in step S2 is a three-layer forward network with a single hidden layer, and the model includes: an input layer, a hidden layer, and an output layer;

[0073] The input layer includes five input neurons for sequentially receiving a first sound velocity parameter, a first maximum amplitude, a first absolute average value, a first root mean square value, and a first waveform index of the transformer oil to be tested;

[0074] The hidden layer includes several hidden neurons, and the transformation function of each hidden neuron is a radial basis function. i For the expression of (X), please refer to formula (6);

[0075] The output layer includes an output neuron for outputting the moisture content prediction result of the transformer oil to be tested.

[0076]

[0077] Where X is the vector composed of the first sound velocity parameter, the first maximum amplitude, the first absolute average value, the first root mean square value and the first waveform index of the transformer oil to be tested, C i is the vector of central parameters of the i-th hidden neuron in the hidden layer, and σ is the standard deviation.

[0078] As an example, the number of hidden neurons included in the hidden layer is 84, and 2σ 2 =1.

[0079] As a preferred solution, the process of obtaining the moisture content prediction model mentioned in step S2 includes steps S21 to S22, and each step is specifically as follows:

[0080] Step S21, performing real-time detection on the transformer oil to be tested in different working conditions to obtain 108 groups of second ultrasonic soundprint features corresponding to the transformer oil to be tested, and using a coulometric trace water meter to perform real-time detection on the transformer oil to be tested in different working conditions to obtain the second moisture content corresponding to each group of second ultrasonic soundprint features; wherein the second ultrasonic soundprint features include a second sound velocity parameter, a second maximum amplitude, a second absolute average value, a second root mean square value, and a second waveform index.

[0081] It should be noted that the process of obtaining the second ultrasonic voiceprint feature is the same as that of obtaining the first ultrasonic voiceprint feature. Please refer to step S1 for details. The experimental process using the coulometric trace water meter includes steps ③ to ⑤, and the details of each step are as follows:

[0082] Step ③: Place the transformer oil to be tested into the coulometric trace water meter, and adjust the agitator rotor until a vortex appears in the electrolyte.

[0083] Step ④: Use a syringe to draw 1 mL of transformer oil to be tested, insert it into the main inlet of the instrument, press the measurement button, and then inject the transformer oil to be tested into the electrolyte.

[0084] Step 5: Repeat the test of the transformer oil under each working condition three times, take the average value, and record the measured data as the second moisture content corresponding to each working condition.

[0085] In step S22, all the second ultrasonic voiceprint features and the second moisture content corresponding to each group of second ultrasonic voiceprint features are taken as a first sample data set.

[0086] Step S23: constructing an RBF neural network model based on the RBF neural network, and using the first sample data set to train and verify the RBF neural network model to obtain a moisture content prediction model.

[0087] As a preferred solution, step S23 includes steps S231 to S233, and the details of each step are as follows:

[0088] Step S231: All the second ultrasonic voiceprint features in the first sample data set and the second moisture content corresponding to each group of second ultrasonic voiceprint features are divided according to a preset division ratio to obtain a training data set and a test data set corresponding to the first sample data set. At this time, the training data set includes 84 groups of second ultrasonic voiceprint features and the second moisture content corresponding to each group of second ultrasonic voiceprint features, and each group of second ultrasonic voiceprint features in the training data set and the second moisture content corresponding to it constitute a training sample, while the test data set includes 24 groups of second ultrasonic voiceprint features and the second moisture content corresponding to each group of second ultrasonic voiceprint features, and each group of second ultrasonic voiceprint features in the test data set and the second moisture content corresponding to it constitute a test sample.

[0089] Step S232: Based on the RBF neural network, an RBF neural network model is constructed, and the RBF neural network model is trained using a training data set to form an initial mapping relationship between ultrasonic voiceprint features and moisture content, thereby obtaining an initial detection model.

[0090] In step S233, each group of second ultrasonic voiceprint features in the test data set is input into the initial detection model to obtain the moisture content prediction results of each second ultrasonic voiceprint feature. The moisture content prediction accuracy of the initial detection model is calculated based on the moisture content prediction results of all second ultrasonic voiceprint features and the second moisture content corresponding to all second ultrasonic voiceprint features. In this embodiment, the moisture content prediction accuracy is characterized by the root mean square error (RMSE), and it is determined whether the RMSE meets the preset range; if so, step S234 is executed; if not, step S235 is executed. The specific calculation method of RMSE is shown in formula (7).

[0091]

[0092] in, is the moisture content prediction result corresponding to the kth test sample in the test data set, y k is the second moisture content corresponding to the kth test sample in the test data set.

[0093] Step S234, completing the training and verification of the RBF neural network model, and using the current initial detection model as the moisture content prediction model;

[0094] Step S235, obtain multiple groups of third ultrasonic voiceprint features corresponding to the transformer oil to be tested under different working conditions, and the third moisture content corresponding to each third ultrasonic voiceprint feature, and use all third ultrasonic voiceprint features and the third moisture content corresponding to each third ultrasonic voiceprint feature as the second sample data set, and iteratively train and verify the current initial detection model based on the current second sample data set to obtain an updated initial detection model, until the moisture content prediction accuracy of the current initial detection model meets the preset range, and then use the current initial detection model as the moisture content prediction model.

[0095] Please refer to Figure 3 , is a schematic diagram of the structure of a transformer oil moisture content prediction system provided by an embodiment of the present invention. The system includes a first data acquisition module M1 and a prediction module M2. The details of each module are as follows:

[0096] A first data acquisition module M1 is used to obtain a first ultrasonic voiceprint feature of the transformer oil to be tested in real time; wherein the first ultrasonic voiceprint feature includes a first sound velocity parameter, a first maximum amplitude, a first absolute average value, a first root mean square value, and a first waveform index;

[0097] The prediction module M2 is used to input the first ultrasonic voiceprint feature into a preset moisture content prediction model, and use the first moisture content output by the moisture content prediction model as the moisture content prediction result of the transformer oil to be tested, so as to complete the moisture content prediction of the transformer oil to be tested; wherein, the moisture content prediction model is obtained by training and verifying the RBF neural network model based on the first sample data set, and the first sample data set includes several groups of second ultrasonic voiceprint features and the second moisture content corresponding to each group of second ultrasonic voiceprint features.

[0098] As a preferred option, please refer to Figure 3 The embodiment of the present invention provides a transformer oil moisture content prediction system, which further includes a second data acquisition module M3 and a model building module M4. The specific details of each module are as follows:

[0099] The second data acquisition module M3 is used to perform real-time detection on the transformer oil to be tested in different working conditions to obtain several groups of second ultrasonic soundprint features corresponding to the transformer oil to be tested, and use a coulometric trace water meter to perform real-time detection on the transformer oil to be tested in different working conditions to obtain the second moisture content corresponding to each group of second ultrasonic soundprint features, and then use all the second ultrasonic soundprint features and the second moisture content corresponding to each group of second ultrasonic soundprint features as the first sample data set; wherein the second ultrasonic soundprint features include a second sound velocity parameter, a second maximum amplitude, a second absolute average value, a second root mean square value, and a second waveform index;

[0100] The model building module M4 is used to build an RBF neural network model based on the RBF neural network, and use the first sample data set to train and verify the RBF neural network model to obtain a moisture content prediction model.

[0101] As a preferred solution, the model building module M4 specifically includes a data partitioning unit 41, a model training unit 42 and a model verification unit 43. The details of each unit are as follows:

[0102] A data partitioning unit 41 is configured to partition all second ultrasonic voiceprint features in the first sample data set and the second moisture content corresponding to each group of second ultrasonic voiceprint features to obtain a training data set and a test data set corresponding to the first sample data set;

[0103] The model training unit 42 is used to construct an RBF neural network model based on the RBF neural network, and train the RBF neural network model using a training data set to form an initial mapping relationship between ultrasonic voiceprint features and moisture content, thereby obtaining an initial detection model;

[0104] The model verification unit 43 is used to input each group of second ultrasonic voiceprint features in the test data set into the initial detection model respectively to obtain the moisture content prediction results of each second ultrasonic voiceprint feature, and calculate the moisture content prediction accuracy of the initial detection model based on the moisture content prediction results of all second ultrasonic voiceprint features and the second moisture content corresponding to all second ultrasonic voiceprint features, and judge whether the moisture content prediction accuracy meets the preset range; if the moisture content prediction accuracy meets the preset range, the training and verification of the RBF neural network model is completed, and the current initial detection model is used as the moisture content prediction model; if the moisture content prediction accuracy does not meet the preset range, multiple groups of third ultrasonic voiceprint features corresponding to the transformer oil to be tested under different working conditions and the third moisture content corresponding to each third ultrasonic voiceprint feature are obtained, and all third ultrasonic voiceprint features and the third moisture content corresponding to each third ultrasonic voiceprint feature are used as the second sample data set, and the current initial detection model is iteratively trained and verified according to the current second sample data set to obtain an updated initial detection model, until the moisture content prediction accuracy of the current initial detection model meets the preset range, and the current initial detection model is used as the moisture content prediction model.

[0105] As a preferred solution, the data acquisition module M1 specifically includes an acquisition and conversion unit 11 and a calculation unit 12. The details of each unit are as follows:

[0106] An acquisition and conversion unit 11 is configured to acquire and convert ultrasonic signals propagating in the transformer oil to be tested in real time using a detection platform to obtain electrical signals corresponding to the ultrasonic signals. The detection platform includes a signal generator, an ultrasonic power amplifier, an oscilloscope, a piezoelectric ultrasonic transmitting transducer, a piezoelectric ultrasonic receiving transducer, and an oil tank.

[0107] The calculation unit 12 is used to display the waveform information corresponding to the electrical signal on the oscilloscope of the detection platform in real time, and calculate the first sound velocity parameter, first maximum amplitude, first absolute average value, first root mean square value and first waveform index corresponding to the transformer oil to be tested according to a preset algorithm and the waveform information.

[0108] As a preferred solution, the RBF neural network model mentioned in the prediction module M2 includes: an input layer, a hidden layer and an output layer;

[0109] The input layer includes five input neurons for sequentially receiving a first sound velocity parameter, a first maximum amplitude, a first absolute average value, a first root mean square value, and a first waveform index of the transformer oil to be tested;

[0110] The hidden layer includes several hidden neurons, and the transformation function of each hidden neuron is a radial basis function;

[0111] The output layer includes an output neuron for outputting the moisture content prediction result of the transformer oil to be tested.

[0112] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0113] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0114] The present invention provides a method and system for predicting the moisture content of transformer oil. Using a moisture content prediction model and combining it with the first ultrasonic soundprint characteristics of the transformer oil to be tested, the system can predict and output the moisture content of the transformer oil to be tested in real time, allowing users to promptly understand the operating status of the transformer and thus avoid potential failure risks. Furthermore, the system uses the first sound velocity parameter, first maximum amplitude, first absolute average value, first root mean square value, and first waveform index as the basis for predicting the moisture content of the transformer oil. This allows the moisture content prediction model to comprehensively analyze both propagation characteristics and time domain characteristics to obtain a moisture content prediction result for the transformer oil to be tested, thereby improving the prediction accuracy of the moisture content prediction model.

[0115] Furthermore, based on the RBF neural network with the best approximation performance and global optimality, an RBF neural network model was constructed and trained and verified to obtain a moisture content prediction model. This optimized the training convergence speed and generalization ability of the moisture content prediction model, thereby improving its training speed and prediction performance. Furthermore, the simple RBF neural network model further accelerated the speed of model training and verification, thereby ensuring the real-time prediction of transformer oil moisture content.

[0116] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for predicting the moisture content of transformer oil, characterized in that: include: Acquire the first ultrasonic soundprint feature of the transformer oil to be tested in real time; The first ultrasonic voiceprint feature is input into a preset moisture content prediction model, and the first moisture content output by the moisture content prediction model is used as the moisture content prediction result of the transformer oil to be tested, so as to complete the moisture content prediction of the transformer oil to be tested; wherein, the acquisition of the moisture content prediction model is specifically as follows: real-time detection is performed on the transformer oil to be tested in different working conditions to obtain several groups of second ultrasonic voiceprint features corresponding to the transformer oil to be tested, and a coulometric trace water meter is used to perform real-time detection on the transformer oil to be tested in different working conditions to obtain the second moisture content corresponding to each group of the second ultrasonic voiceprint features, and then all the second ultrasonic voiceprint features and the second moisture content corresponding to each group of the second ultrasonic voiceprint features are used as the first sample data set; based on the RBF neural network, an RBF neural network model is constructed, and the RBF neural network model is trained and verified using the first sample data set to obtain the moisture content prediction model; wherein, the second ultrasonic voiceprint feature includes a second sound velocity parameter, a second maximum amplitude, a second absolute average value, a second root mean square value, and a second waveform index; Among them, the first ultrasonic voiceprint feature includes a first sound speed parameter, a first maximum amplitude, a first absolute average value, a first root mean square value and a first waveform index. The moisture content prediction model is obtained by training and verifying the RBF neural network model based on the first sample data set. The first sample data set includes several groups of the second ultrasonic voiceprint features and the second moisture content corresponding to each group of the second ultrasonic voiceprint features.

2. The method for predicting moisture content of transformer oil according to claim 1, wherein: The RBF neural network model is constructed based on the RBF neural network, and the RBF neural network model is trained and verified using the first sample data set to obtain the moisture content prediction model, specifically: Dividing all the second ultrasonic voiceprint features in the first sample data set and the second moisture content corresponding to each group of the second ultrasonic voiceprint features to obtain a training data set and a test data set corresponding to the first sample data set; Based on the RBF neural network, the RBF neural network model is constructed, and the RBF neural network model is trained using the training data set to form an initial mapping relationship between ultrasonic voiceprint features and moisture content, thereby obtaining an initial detection model; Inputting each group of the second ultrasonic voiceprint features in the test data set into the initial detection model to obtain a moisture content prediction result for each second ultrasonic voiceprint feature, and calculating the moisture content prediction accuracy of the initial detection model based on the moisture content prediction results of all the second ultrasonic voiceprint features and the second moisture contents corresponding to all the second ultrasonic voiceprint features, and determining whether the moisture content prediction accuracy meets a preset range; If so, the training and verification of the RBF neural network model is completed, and the current initial detection model is used as the moisture content prediction model; If not, obtain multiple groups of third ultrasonic voiceprint features corresponding to the transformer oil to be tested under different working conditions, and the third moisture content corresponding to each of the third ultrasonic voiceprint features, and use all the third ultrasonic voiceprint features and the third moisture content corresponding to each of the third ultrasonic voiceprint features as the second sample data set, and iteratively train and verify the current initial detection model based on the current second sample data set to obtain the updated initial detection model, until the moisture content prediction accuracy of the current initial detection model meets the preset range, and then use the current initial detection model as the moisture content prediction model.

3. The method for predicting moisture content of transformer oil according to claim 1, wherein: The real-time acquisition of the first ultrasonic voiceprint feature of the transformer oil to be tested is specifically as follows: Using a detection platform, the ultrasonic signal propagating in the transformer oil to be tested is collected and converted in real time to obtain an electrical signal corresponding to the ultrasonic signal; On the oscilloscope of the detection platform, waveform information corresponding to the electrical signal is displayed in real time, and according to a preset algorithm, the first sound velocity parameter, the first maximum amplitude, the first absolute average value, the first root mean square value, and the first waveform index corresponding to the transformer oil to be tested are calculated in combination with the waveform information; The detection platform includes a signal generator, an ultrasonic power amplifier, the oscilloscope, a piezoelectric ultrasonic transmitting transducer, a piezoelectric ultrasonic receiving transducer and a fuel tank.

4. The method for predicting moisture content of transformer oil according to claim 1, wherein: The RBF neural network model includes: an input layer, a hidden layer and an output layer; The input layer includes five input neurons for sequentially receiving the first sound velocity parameter, the first maximum amplitude, the first absolute average value, the first root mean square value, and the first waveform index of the transformer oil to be tested; The hidden layer includes a plurality of hidden neurons, and the transformation function of each hidden neuron is a radial basis function; The output layer includes an output neuron, which is used to output the moisture content prediction result of the transformer oil to be tested.

5. A system for predicting moisture content of transformer oil, characterized in that: include: A first data acquisition module is configured to acquire, in real time, a first ultrasonic voiceprint feature of the transformer oil to be tested; wherein the first ultrasonic voiceprint feature includes a first sound velocity parameter, a first maximum amplitude, a first absolute average value, a first root mean square value, and a first waveform index; A prediction module, configured to input the first ultrasonic voiceprint feature into a preset moisture content prediction model, and use the first moisture content output by the moisture content prediction model as a moisture content prediction result of the transformer oil to be tested, so as to complete the moisture content prediction of the transformer oil to be tested; wherein the moisture content prediction model is obtained by training and verifying an RBF neural network model based on a first sample data set, and the first sample data set includes several groups of second ultrasonic voiceprint features and the second moisture content corresponding to each group of the second ultrasonic voiceprint features; wherein the moisture content prediction model is obtained by: performing real-time detection on the transformer oil to be tested in different working conditions to obtain several groups of the second ultrasonic voiceprint features corresponding to the transformer oil to be tested, and using a coulometric trace water meter to perform real-time detection on the transformer oil to be tested in different working conditions to obtain the second moisture content corresponding to each group of the second ultrasonic voiceprint features, and then using all the second ultrasonic voiceprint features and the second moisture content corresponding to each group of the second ultrasonic voiceprint features as the first sample data set; A model construction module is used to construct the RBF neural network model based on the RBF neural network, and use the first sample data set to train and verify the RBF neural network model to obtain the moisture content prediction model; wherein, the second ultrasonic voiceprint feature includes a second sound speed parameter, a second maximum amplitude, a second absolute average value, a second root mean square value and a second waveform index.

6. A transformer oil moisture content prediction system according to claim 5, characterized in that: The model building module specifically includes: a data partitioning unit, configured to partition all the second ultrasonic voiceprint features in the first sample data set, and the second moisture content corresponding to each group of the second ultrasonic voiceprint features, to obtain a training data set and a test data set corresponding to the first sample data set; A model training unit is used to construct the RBF neural network model based on the RBF neural network, and train the RBF neural network model using the training data set to form an initial mapping relationship between ultrasonic voiceprint features and moisture content, thereby obtaining an initial detection model; A model verification unit is used to input each group of the second ultrasonic voiceprint features in the test data set into the initial detection model respectively to obtain the moisture content prediction results of each second ultrasonic voiceprint feature, and calculate the moisture content prediction accuracy of the initial detection model according to the moisture content prediction results of all the second ultrasonic voiceprint features and the second moisture content corresponding to all the second ultrasonic voiceprint features, and judge whether the moisture content prediction accuracy meets the preset range; if the moisture content prediction accuracy meets the preset range, the training verification of the RBF neural network model is completed, and the current initial detection model is used as the moisture content prediction model; if If the moisture content prediction accuracy does not meet the preset range, multiple groups of third ultrasonic voiceprint features corresponding to the transformer oil to be tested under different working conditions and the third moisture content corresponding to each of the third ultrasonic voiceprint features are obtained, and all the third ultrasonic voiceprint features and the third moisture content corresponding to each of the third ultrasonic voiceprint features are used as the second sample data set, and the current initial detection model is iteratively trained and verified based on the current second sample data set to obtain the updated initial detection model, until the moisture content prediction accuracy of the current initial detection model meets the preset range, and the current initial detection model is used as the moisture content prediction model.

7. A transformer oil moisture content prediction system according to claim 5, characterized in that: The first data acquisition module specifically includes: An acquisition and conversion unit is used to acquire and convert the ultrasonic signal propagating in the transformer oil to be tested in real time using a detection platform to obtain an electrical signal corresponding to the ultrasonic signal; wherein the detection platform includes a signal generator, an ultrasonic power amplifier, an oscilloscope, a piezoelectric ultrasonic transmitting transducer, a piezoelectric ultrasonic receiving transducer, and an oil tank; a computing unit configured to display waveform information corresponding to the electrical signal on the oscilloscope of the detection platform in real time, and to calculate, in accordance with a preset algorithm and in combination with the waveform information, the first sound velocity parameter, the first maximum amplitude, the first absolute average value, the first root mean square value, and the first waveform index corresponding to the transformer oil to be tested.

8. A transformer oil moisture content prediction system according to claim 5, characterized in that: The RBF neural network model includes: an input layer, a hidden layer and an output layer; The input layer includes five input neurons for sequentially receiving the first sound velocity parameter, the first maximum amplitude, the first absolute average value, the first root mean square value, and the first waveform index of the transformer oil to be tested; The hidden layer includes a plurality of hidden neurons, and the transformation function of each hidden neuron is a radial basis function; The output layer includes an output neuron, which is used to output the moisture content prediction result of the transformer oil to be tested.

Images