A High-Precision Detection Method for Trace Gases in Transformer Oil Based on Error Compensation
By analyzing sensor electrical signals and environmental data in extremely cold environments and screening for significant interference moments for baseline correction, the error compensation problem of trace gas detection in transformer oil under low-temperature conditions was solved, and high-precision concentration detection was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEILONGJIANG ELECTRIC POWER SCIENCE RESEARCH INSTITUTE
- Filing Date
- 2026-05-15
- Publication Date
- 2026-07-17
AI Technical Summary
In extremely cold environments, the detection signal of trace gases in transformer oil is easily affected by electromagnetic interference and mechanical vibration, resulting in insufficient error compensation and incomplete signal correction of traditional detection methods, thus reducing the accuracy of concentration inversion.
By collecting electrical signals and environmental data from sensors in low-temperature environments, analyzing the stage interference and outlier degree of the environmental data, screening out moments with significant interference, obtaining the comprehensive influence coefficient, performing baseline offset correction, and combining noise reduction processing and neural network models to achieve error compensation.
It improves the accuracy of detecting trace gas concentrations in transformer oil under low-temperature conditions, with an error of less than 2 ppm and a long-term drift rate of less than 3%, ensuring the authenticity and stability of the detection results.
Smart Images
Figure CN122218203B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of trace gas detection technology, specifically to a high-precision detection method for trace gases in transformer oil based on error compensation. Background Technology
[0002] Detecting trace gases in transformer oil is a core technology for ensuring the safe and stable operation of large power equipment during transformer operation. During long-term operation, the internal insulation material of a transformer may decompose due to faults such as partial discharge, overheating, and arcing, producing gases that dissolve in the insulating oil. and Characteristic gases, such as those found in transformer oil, have concentrations that directly reflect the type and severity of internal transformer faults. Therefore, accurate detection of trace gas concentrations in transformer oil can precisely determine the transformer's health status, prevent sudden faults and power outages, and ensure the stable operation of the power grid.
[0003] However, in actual detection processes, the extreme cold environment can significantly reduce the carrier migration efficiency of nano-gas-sensitive materials. At the same time, electromagnetic interference and mechanical vibration can also cause sensor baseline drift and sensitivity decay, which may lead to distortion of the detection signal. Traditional trace gas concentration detection algorithms are not capable of handling the nonlinear error of the detection signal caused by the superposition of multiple factors, resulting in insufficient error compensation and incomplete signal correction, which greatly reduces the accuracy of concentration inversion and causes the detection results to deviate from the true value. Summary of the Invention
[0004] In view of the above, it is necessary to provide a high-precision detection method for trace gases in transformer oil based on error compensation. Compared with the traditional high-precision detection method for trace gases in transformer oil, the detection accuracy of trace gas concentration in transformer oil under low temperature environment is improved by improving the error compensation accuracy.
[0005] The high-precision detection method for trace gases in transformer oil based on error compensation in this application adopts the following technical solution:
[0006] One embodiment of this application provides a high-precision detection method for trace gases in transformer oil based on error compensation. The method includes the following steps:
[0007] Within a preset time period under a preset temperature environment, the dynamic response curves of various electrical signals generated by the sensor when detecting the concentration of trace gases in transformer oil are collected, as well as various environmental data at each time point. The environmental data includes ambient temperature, sensor surface temperature, sensor vibration amplitude, and substation electromagnetic interference intensity.
[0008] By analyzing the fluctuations of individual environmental data, the data is divided into different time intervals. By comparing the number of individual environmental data in each time interval with the preset time period, and the intensity of fluctuations of individual environmental data in each time interval, the stage interference degree of individual environmental data is obtained.
[0009] The outlier degree of various environmental data at each time point is measured and combined with the stage interference degree of all environmental data to obtain the comprehensive interference significance at each time point, and then the time point with significant interference is screened out.
[0010] By identifying significant interference moments and their nearest time ranges, we obtain the characteristic analysis time periods. By analyzing the characteristic deviation of the dynamic response curves of various electrical signals relative to the baseline response curve within each characteristic analysis time period, we obtain the comprehensive influence coefficient of each characteristic analysis time period. This coefficient is then used to correct the theoretical baseline deviations obtained using the preset temperature drift model, resulting in the actual baseline deviations for each characteristic analysis time period. These deviations are then used to synchronously correct the dynamic response curves of various electrical signals in order to detect the concentration of trace gases in transformer oil. The process for obtaining the stage interference degree is as follows: Calculate the proportion of a single type of environmental data in each time interval among all single type of environmental data in the preset time period; Calculate the dispersion of the normalized values of a single environmental data type within each time interval; The stage interference degree is obtained by combining the proportion of the quantity and the dispersion across all time intervals.
[0011] The process for obtaining the comprehensive interference significance is as follows: The product of the normalized value of the stage interference of various environmental data and the measure of the outlier degree of various environmental data at each time point is used as the significance of the influence of various environmental data at each time point. The comprehensive interference significance is the sum of the influence significance of all environmental data at each time point.
[0012] In one embodiment, the calculation process for the stage interference degree is as follows:
[0013] The stage characteristic values of a single type of environmental data in each time interval are obtained by the proportion of the quantity and the dispersion of the single type of environmental data in each time interval, and the stage characteristic values are positively correlated with the proportion of the quantity and the dispersion, respectively.
[0014] The stage interference degree is the sum of the stage characteristic values of a single type of environmental data across all time intervals.
[0015] In one embodiment, the moment when the interference becomes significant is the moment when the overall interference significance is greater than a preset threshold.
[0016] In one embodiment, the process of obtaining the comprehensive influence coefficient is as follows:
[0017] For a single feature analysis time period, the preset curve feature values of the dynamic response curve and the reference response curve of various electrical signals are extracted respectively, and the difference of all curve feature values between the dynamic response curve of various electrical signals and the reference response curve is comprehensively measured.
[0018] The comprehensive influence coefficient for a single feature analysis time period is obtained by using the difference between the dynamic response curves of all types of electrical signals and their reference response curves.
[0019] In one embodiment, the comprehensive influence coefficient of the single feature analysis time period is the weighted sum of the normalized values of the differences between the dynamic response curves of all electrical signals and their reference response curves.
[0020] In one embodiment, the actual baseline offset is the product of the comprehensive influence coefficient and the theoretical baseline offset.
[0021] In one embodiment, during the synchronous correction of the dynamic response curves of various electrical signals, the time not within any feature analysis time period is recorded as a stable time. The theoretical baseline offset is obtained using a preset temperature drift model. The theoretical baseline offset is used as the actual baseline offset to correct the baseline offset of the stable time. The time within two or more feature analysis time periods is recorded as a coincident time. From the center time of all feature analysis time periods where the coincident time is located, the center time with the shortest time interval between it and the coincident time is selected. The coincident time is then corrected according to the actual baseline offset corresponding to the selected center time.
[0022] This application has at least the following beneficial effects:
[0023] This application analyzes the temporal fluctuation characteristics of single environmental data and divides time intervals to accurately capture the phased fluctuation patterns of single environmental data during concentration detection. By comparing the number of data in each time interval with that in a preset time period, it reflects the continuous impact range of the fluctuation changes of single environmental data. By calculating the dispersion in each time interval, it quantifies the intensity of environmental data fluctuations. Furthermore, by combining the proportion of data and the dispersion, it obtains the phased interference degree, which can comprehensively assess the significance of the impact of single environmental data on the sensor in different time intervals, achieve a phased and refined assessment of environmental interference analysis, and provide a single-factor reference for subsequently determining the comprehensive interference level at each moment.
[0024] Furthermore, by measuring the outlier degree of various environmental data at different times, it is possible to accurately identify the moments when there are abnormal environmental disturbances. This can then be combined with the stage interference degree to obtain the comprehensive interference significance, taking into account both the stage fluctuation characteristics of environmental data and the degree of deviation of single-point anomalies, and achieving a comprehensive quantitative assessment of multiple environmental interference factors. In addition, the significant interference moments can be screened to determine the feature analysis time period, which can accurately locate the time period when the sensor is most severely affected by comprehensive interference.
[0025] Furthermore, within each characteristic analysis time period, by measuring the characteristic deviation of the dynamic response curves of various electrical signals relative to the reference response curve, a reliable basis is provided for the subsequent dynamic adjustment of the baseline offset, so that the baseline correction intensity is dynamically matched with the actual interference level, improving the error compensation accuracy, enhancing the authenticity and stability of the dynamic response curve, and thus improving the detection accuracy of trace gas concentration in transformer oil under low temperature conditions. Attached Figure Description
[0026] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0027] Figure 1 A flowchart illustrating the steps of the high-precision detection method for trace gases in transformer oil based on error compensation provided in this application;
[0028] Figure 2 This is a schematic diagram of the screening process for moments of significant interference.
[0029] Figure 3 This is a schematic diagram of the calibration process for the dynamic response curve. Detailed Implementation
[0030] In the description of the embodiments in this application, the words "exemplary," "or," and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary," "or," and "for example" is intended to present the relevant concepts in a specific manner.
[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. It should be understood that, unless otherwise stated, " / " in this application means "or".
[0032] It should also be noted that the terms "first" and "second" in this application are used to distinguish similar objects, rather than to describe a specific order or sequence.
[0033] The following description, in conjunction with the accompanying drawings, details the specific scheme of the high-precision detection method for trace gases in transformer oil based on error compensation provided in this application.
[0034] This application provides an embodiment of a high-precision detection method for trace gases in transformer oil based on error compensation. Specifically, the following high-precision detection method for trace gases in transformer oil based on error compensation is provided. Please refer to [link to relevant documentation]. Figure 1 The method includes the following steps:
[0035] Step 1: Within a preset time period under a preset temperature environment, collect the dynamic response curves of various electrical signals generated by the sensor when detecting the concentration of trace gases in transformer oil, as well as various environmental data at each time point. The environmental data includes ambient temperature, sensor surface temperature, sensor vibration amplitude, and substation electromagnetic interference intensity.
[0036] In the process of detecting trace gas concentrations in transformer oil under low-temperature conditions, the low temperature environment leads to a decrease in the surface activity of the nano-gas-sensitive material, a significant drop in carrier migration efficiency, increased baseline drift of the sensor, and a significant decrease in sensitivity. At the same time, the oil-gas separation efficiency decreases and pipeline adsorption increases, and the weak gas response signal may be submerged by environmental noise and temperature drift. Therefore, in order to suppress low-temperature interference, ensure that the sensor operates within the optimal temperature range, and achieve stable acquisition of weak signals and synchronous acquisition of multi-source error data, this application uses a constant-temperature integrated low-power micro-sensor module and a multi-channel synchronous data acquisition unit for data acquisition. The constant-temperature integrated low-power micro-sensor module mainly consists of an extreme cold-specific nano-semiconductor gas-sensitive sensor, a low-power micro-constant-temperature control unit, a high-efficiency oil-gas separation membrane assembly, an inert gas flow channel, an anti-electromagnetic interference packaging shell, and a real-time temperature monitoring chip. The multi-channel synchronous data acquisition unit consists of a 16-bit high-precision ADC (Analog to Digital Converter) sampling module, a multi-channel signal conditioning circuit, an environmental parameter acquisition interface, a synchronous timing controller, an embedded data processing core, and a storage and communication interface. During data acquisition, the low-power miniature temperature-controlled unit stably controls the core working area of the sensor at a constant temperature. The high-efficiency oil-gas separation membrane module has a gas flow rate of 5 mL / min to 10 mL / min, and the 16-bit high-precision ADC sampling module has a sampling frequency of 1 Hz. All channels adopt a synchronous sampling mode to align the weak gas response signal with the environmental disturbance signal in the time dimension, thereby reducing the impact on the accuracy of subsequent error analysis and compensation. It should be noted that due to the heating power limit in low-temperature environments and the uneven thermal field distribution of the sensor module, the core sensitive area of the sensor still has residual temperature drift with environmental fluctuations. The 1 Hz setting is merely one embodiment of this application; implementers can set it according to actual conditions, and this application does not impose any special limitations.
[0037] In this embodiment, the gas flow rate of the high-efficiency oil-gas separation membrane module is 5 mL / min.
[0038] Specifically, within a preset time period under a preset temperature environment, the system collects dynamic response curves of various electrical signals generated by the sensor when detecting the concentration of trace gases in transformer oil, as well as various environmental data at each moment. The sampled values on the dynamic response curves and the environmental data are stored synchronously with timestamps for subsequent data feature analysis and high-precision error compensation.
[0039] In this embodiment, the preset temperature range of the low-temperature environment is from minus forty degrees Celsius to zero degrees Celsius, and the preset temperature is minus ten degrees Celsius.
[0040] In this embodiment, the length of the preset time period is 30 minutes. The length of the preset time period is preset by the user and the implementer can set it according to the actual situation. This application does not impose any special restrictions.
[0041] In this embodiment, the dynamic response curve is specifically: the sensor collects the data on the dissolved substances in the transformer oil. , The dynamic response curves of various electrical signals generated during concentration detection, including the dynamic response curves of voltage signals and current signals.
[0042] In this embodiment, the environmental data includes ambient temperature, sensor surface temperature, sensor vibration amplitude, and substation electromagnetic interference intensity.
[0043] Step 2: Obtain the stage interference degree of single environmental data; obtain the comprehensive interference significance at each time point, and then filter out the time points with significant interference; obtain the time periods for each feature analysis, obtain the comprehensive influence coefficient of each feature analysis time period, obtain the actual baseline offset of each feature analysis time period, and synchronously correct the dynamic response curves of various electrical signals.
[0044] In the operating environment of transformers, strong electromagnetic interference, mechanical vibration, and current fluctuations generated by high-voltage equipment can introduce a large amount of high-frequency random noise and pulse interference into the sensor signal. This may directly mask the weak response signal generated by ppb-level trace gases, resulting in a significant reduction in the signal-to-noise ratio and affecting the accuracy of subsequent feature extraction and trace gas concentration calculation. Therefore, in order to eliminate noise interference, this application performs noise reduction processing on the dynamic response curve within the preset time period, eliminating high-frequency noise and abnormal pulses. While retaining the true signal peak and trend, it improves the signal-to-noise ratio, providing an accurate basic signal for subsequent error compensation and concentration inversion.
[0045] In this embodiment, a wavelet filtering algorithm is used for noise reduction. The wavelet filtering algorithm is a well-known technology and will not be described in detail here. As other implementation methods, based on the ability to reduce noise in the dynamic response curve, implementers may use other existing feasible technologies. This application does not impose any special restrictions.
[0046] Furthermore, during the detection of trace gases in transformer oil at low temperatures, changes in ambient temperature and fluctuations in sensor operating temperature alter the surface activity and carrier migration rate of the nano-sensitive materials. Additionally, electromagnetic interference and mechanical vibration can also affect the sensor, causing nonlinear and strongly coupled drift in the sensor baseline and sensitivity, leading to systematic deviations in the trace gas concentration detection results. Therefore, based on the above analysis, considering the significant temporal differences, stage fluctuations, and coupled superposition of the influence of temperature and other interference factors in the actual processing at low temperatures, and the varying degrees of influence of different interferences at the same time, resulting in alternating high and low non-stationary changes in the overall interference intensity over different time periods, it is impossible to achieve accurate correction throughout the entire time period using a uniform and fixed compensation parameter. Therefore, it is necessary to analyze the sensor offset characteristics under different levels of overall interference to accurately compensate for the sensor baseline offset under low-temperature conditions.
[0047] Step 2.1: Based on the fluctuation of individual environmental data, divide the individual environmental data into various time intervals. By comparing the number of individual environmental data in each time interval with the preset time period, and the intensity of fluctuation of individual environmental data in each time interval, obtain the stage interference degree of individual environmental data. Measure the outlier degree of various environmental data at each time point, and combine it with the stage interference degree of all environmental data to obtain the comprehensive interference significance at each time point, and then filter out the time points with significant interference.
[0048] Since the overall interference intensity varies in different time periods, this application analyzes the temporal fluctuation characteristics of environmental data and divides time intervals based on the analysis results in order to subsequently determine the periods when the sensor is significantly affected by interference.
[0049] By analyzing the fluctuations of individual environmental data, the data is divided into different time intervals to identify the phased fluctuation characteristics of individual environmental data during concentration detection, thereby determining the dynamic fluctuation impact characteristics under the influence of individual environmental interference factors.
[0050] In this embodiment, the BG (Bernaola Galvan) segmentation algorithm is used to detect the abrupt changes in the temporal sequence of a single environmental data within the preset time period. The time of the abrupt change is taken as the segmentation time, and the preset time period is divided into time intervals. The BG segmentation algorithm is a well-known technology and will not be described in detail in this application. As other implementation methods, based on the ability to detect the abrupt changes in the temporal sequence of a single environmental data within the preset time period, implementers may use other existing feasible technologies, and this application does not impose any special restrictions.
[0051] Furthermore, by comparing the number of individual environmental data points within each time interval with the preset time period, and the degree of fluctuation of individual environmental data within each time interval, the stage interference degree of individual environmental data is obtained. The specific process is as follows:
[0052] Calculate the proportion of a single environmental data point within each time interval to the total number of single environmental data points within the preset time period, in order to reflect the characteristics of the continuous impact range of fluctuations in the single environmental data point.
[0053] Calculate the dispersion of the normalized values of a single environmental data point within each time interval to reflect the degree of fluctuation of the single environmental data point within each time interval.
[0054] The stage characteristic values of a single type of environmental data in each time interval are obtained by the proportion of the quantity and the dispersion of the single type of environmental data in each time interval, and the stage characteristic values are positively correlated with the proportion of the quantity and the dispersion, respectively.
[0055] The stage interference degree of a single environmental data is the sum of the stage characteristic values of the single environmental data across all time intervals.
[0056] It should be noted that positive correlation means that the variables change in the same direction; when one variable increases, the other variable also increases, and when one variable decreases, the other variable also decreases.
[0057] In this embodiment, the dispersion is specifically the coefficient of variation. The calculation of the coefficient of variation is a well-known technique and will not be described in detail here. As other implementation methods, based on the ability to measure the normalized value of a single environmental data, the implementer may adopt other existing feasible techniques, such as variance, standard deviation, etc. This application does not impose any special restrictions.
[0058] In this embodiment, the Min-Max normalization method is used to obtain the normalized value of a single environmental data. When normalizing the single environmental data, the maximum value refers to the maximum value of the single environmental data within the preset time period, and the minimum value refers to the minimum value of the single environmental data within the preset time period. The Min-Max normalization method is a well-known technology and will not be described in detail in this application.
[0059] In this embodiment, the expression for the stage characteristic value of a single type of environmental data in each time interval is:
[0060] In the formula, This represents the stage feature value of the x-th type of environmental data in the i-th time interval; This represents the proportion of the x-th type of environmental data within the i-th time interval; This represents the dispersion of the x-th type of environmental data within the i-th time interval; This indicates a preset value greater than 0, used to avoid... When it is 0, The case where it is forced to be 0. Among them, The value is 0.01. The value is preset by the user, and the implementer can set it according to the actual situation. This application does not impose any special restrictions.
[0061] It should be noted that the greater the calculated stage interference, the more significant the stage changes of a single environmental data point during concentration detection have on the sensor, and the higher the degree of influence of a single environmental data point on the subsequent baseline correction time period division.
[0062] Furthermore, the outlier degree of various environmental data at each time point is measured and combined with the stage interference degree of all environmental data to obtain the comprehensive interference significance at each time point. The specific process is as follows:
[0063] By measuring the stage interference of various environmental data and the outlier degree of various environmental data at each time point, the significance of the influence of various environmental data at each time point is obtained. The significance of the influence is positively correlated with the stage interference of various environmental data and also positively correlated with the measurement result of the outlier degree of various environmental data at each time point.
[0064] The overall significance of interference at each time point is the sum of the significance of the influence of all environmental data at each time point.
[0065] The product of the normalized value of the stage interference degree of various environmental data and the measurement result of the outlier degree of various environmental data at each time point is used as the significance of the influence of various environmental data at each time point.
[0066] In this embodiment, the stage interference degree of all kinds of environmental data is used as input, and the Softmax function is used to output the normalized value of the stage interference degree of various environmental data. The Softmax function is a well-known technology and will not be described in detail in this application.
[0067] In this embodiment, the method for measuring the outlier degree of various environmental data at each time point is as follows: various environmental data within the preset time period are used as inputs to the Local Outlier Factor (LOF) algorithm, and the LOF values of various environmental data at each time point are output to reflect the significance of the influence of various environmental data on the sensor at each time point. The LOF algorithm is a well-known technology and will not be described in detail in this application. As other implementation methods, based on the ability to measure the outlier degree of various environmental data at each time point, implementers may adopt other existing feasible technologies, and this application does not impose any special restrictions.
[0068] It should be noted that the greater the calculated significance of the overall interference, the more significant the combined impact of various environmental interference factors on the sensor at each time point.
[0069] Furthermore, by analyzing the comprehensive interference salience at each moment within the preset time period, moments with significant interference are selected, specifically as follows:
[0070] The moment when the overall interference significance exceeds a preset threshold is defined as the moment when the interference is significant. A schematic diagram of the filtering process for moments of significant interference is shown below. Figure 2 As shown.
[0071] In this embodiment, the method for obtaining the preset threshold is as follows: by using the comprehensive interference significance of all collection times during the historical 50 concentration detections, the 3Sigma anomaly detection algorithm is used to obtain the 3Sigma range of the comprehensive interference significance of all collection times during the historical 50 concentration detections, and the upper limit of the 3Sigma range is used as the preset threshold. The 3Sigma anomaly detection algorithm is a well-known technology and will not be described in detail in this application. 50 is just one embodiment of this application, and implementers can set it according to the actual situation. This application does not impose any special restrictions.
[0072] Step 2.2: Obtain the characteristic analysis time period by the significant time of each interference and its nearest time range. Obtain the comprehensive influence coefficient of each characteristic analysis time period by the characteristic deviation of the dynamic response curve of various electrical signals relative to the reference response curve within each characteristic analysis time period.
[0073] By identifying the significant moments of interference and their nearest neighboring time ranges, the time periods for each feature analysis are obtained, specifically:
[0074] Centered on each significant moment of interference, we trace back and extend forward to obtain a pre-defined complete gas detection cycle time span centered on each significant moment of interference, which serves as the time period for each feature analysis.
[0075] In this embodiment, the length of the complete gas detection cycle is 10 minutes, which is 5 minutes backward and 5 minutes forward. The length of the complete gas detection cycle is determined based on the physical duration of oil-gas separation and gas-sensitive response in the temperature-controlled integrated low-power micro-sensor module.
[0076] Furthermore, by analyzing the degree of characteristic deviation of the dynamic response curves of various electrical signals relative to the reference response curve within each characteristic analysis time period, the comprehensive influence coefficient of each characteristic analysis time period is obtained. This coefficient reflects the degree of characteristic deviation under the influence of comprehensive interference within each characteristic analysis time period. The specific process is as follows:
[0077] For a single feature analysis time period, the preset curve feature values of the dynamic response curves and the reference response curve of various electrical signals are extracted respectively. The difference of all curve feature values between the dynamic response curves of various electrical signals and the reference response curve is comprehensively measured to quantify the feature offset and distortion caused by environmental interference. The reference response curve is the dynamic response curve collected under an ideal state of constant temperature and electromagnetic shielding without interference.
[0078] The expression for the comprehensive influence coefficient of each feature analysis time period is as follows:
[0079] In the formula, This represents the overall impact coefficient for a single feature analysis period. The normalized value representing the difference between all characteristic values of the current signal's dynamic response curve and its reference response curve. The normalized value representing the difference between all characteristic values of the dynamic response curve of a voltage signal and its reference response curve. , All represent preset weights greater than 0, used to reflect the contribution of current signals and voltage signals, as well as monitoring reliability, respectively.
[0080] In this embodiment, the curve feature values include the rise rate, decay time constant, and signal fluctuation index. The signal fluctuation index is obtained by calculating the variance of the first difference of the amplitudes of all adjacent sampling points on the curve, and is used to quantify the irregularity of the waveform. Implementers can add other feasible curve feature values according to the actual situation, and this application does not impose any special restrictions.
[0081] In this embodiment, the process of measuring the difference between all curve characteristic values of the dynamic response curves of various electrical signals and their reference response curves is as follows: The Min-Max normalization method is used to normalize each curve characteristic value. The normalized values of all curve characteristic values of the dynamic response curves of various electrical signals are used to form the response characteristic vectors of the dynamic response curves of various electrical signals. The normalized values of all curve characteristic values of the reference response curves of various electrical signals are used to form the response characteristic vectors of the reference response curves of various electrical signals. The Euclidean distance between the response characteristic vectors of the dynamic response curves of various electrical signals and their reference response curves is calculated. The calculation of the Euclidean distance is a well-known technique and will not be elaborated upon in this application. As other implementation methods, based on the ability to measure the degree of difference between curve characteristic values, implementers can adopt other existing feasible techniques, and this application does not impose any special restrictions. Taking current signals and signal fluctuation indices as examples, when normalizing the signal fluctuation index, the maximum value refers to the maximum value among the signal fluctuation indices of the dynamic response curves of the current signals and their reference response curves within all feature analysis time periods, and the minimum value refers to the minimum value among the signal fluctuation indices of the dynamic response curves of the current signals and their reference response curves within all feature analysis time periods.
[0082] In this embodiment, the Min-Max normalization method is used to obtain the normalized value of the difference. When normalizing the difference, taking the current signal as an example, the maximum value refers to the maximum value of the difference between the dynamic response curve of the current signal and its reference response curve in all feature analysis time periods, and the minimum value refers to the minimum value of the difference between the dynamic response curve of the current signal and its reference response curve in all feature analysis time periods.
[0083] In this embodiment, and The value is 0.5 for all values, taking into account the interference response characteristics of the two electrical signals, thus improving the balance and stability of the comprehensive influence coefficient calculation. and The values are all preset by humans, and the implementer can determine the specific values according to the sensor type, gas sensing mechanism, signal output characteristics and on-site calibration test results.
[0084] It should be noted that the larger the calculated comprehensive influence coefficient, the greater the degree of feature shift under the influence of comprehensive interference within each feature analysis time period.
[0085] Step 2.3: By using the comprehensive influence coefficient of each feature analysis time period, correct the theoretical offset of each baseline obtained by the preset temperature drift model, and obtain the actual offset of the baseline for each feature analysis time period, which is used to synchronously correct the dynamic response curves of various electrical signals.
[0086] The average ambient temperature and average sensor surface temperature within each feature analysis time period are used as inputs. A temperature drift model is employed to output the theoretical baseline offset for each feature analysis time period. The theoretical baseline offset is adjusted based on the comprehensive influence coefficient of each feature analysis time period to match the baseline correction strength with the actual interference level, avoiding over-correction during low-interference periods and under-compensation during high-interference periods.
[0087] In this embodiment, the temperature drift model is specifically a baseline drift nonlinear fitting model. The specific process of using the temperature drift model for quantitative calculation and analysis is a technique known to those skilled in the art, and will not be described in detail here.
[0088] Based on the above analysis, the theoretical baseline offset obtained using the temperature drift model is corrected by using the comprehensive influence coefficient of each feature analysis time period, resulting in the actual baseline offset for each feature analysis time period, expressed as:
[0089] Where P represents the actual baseline offset during a single feature analysis time period; This represents the baseline theoretical offset for a single feature analysis time period; This represents the comprehensive impact coefficient for a single feature analysis period.
[0090] It should be noted that: when the comprehensive influence coefficient The closer the value is to 0, the closer the dynamic response curve of the electrical signal is to its baseline response curve within a single feature analysis period. This indicates a weaker degree of interference affecting the dynamic response curve of the electrical signal, and the closer the actual baseline offset is to 0. Only a very small correction is performed, achieving fine-tuning under low interference. When the comprehensive influence coefficient... The closer the value is to 1, the greater the interference affecting the dynamic response curve of the electrical signal within a single feature analysis period. The greater the distortion of the dynamic response curve of the electrical signal may be, the more the actual baseline offset is amplified synchronously, and the compensation strength automatically increases with the increase of the interference level, so that the compensation amplitude is positively correlated with the comprehensive interference intensity, so as to achieve accurate baseline offset correction.
[0091] Furthermore, the calculated actual baseline offset for each characteristic analysis time period is used as a correction value and substituted point by point into the dynamic response curve. Using the timestamp as a matching reference, the dynamic response curves of the current signal and the voltage signal are synchronously corrected. The baseline offset component is subtracted in real time within each characteristic analysis time period to eliminate signal reference deviation caused by multi-source interference. This effectively removes systematic errors caused by temperature, vibration, and electromagnetic interference coupling in low-temperature environments, improving the authenticity and stability of the dynamic response curve. A schematic diagram of the dynamic response curve correction process is shown below. Figure 3 As shown.
[0092] It should be added that, during the synchronous correction of the dynamic response curves of various electrical signals, since the division of the feature analysis time period is determined based on the physical duration of oil-gas separation and gas-sensitive response in the isothermal integrated low-power micro sensor module, after the feature analysis time period is divided, the moments not within any feature analysis time period correspond to a stable state with stable interference and no obvious signal distortion. These moments are recorded as stable moments. The theoretical baseline offset is obtained using a temperature drift model based on the ambient temperature and sensor surface temperature at the stable moment. The theoretical baseline offset is then used as the actual baseline offset to correct the baseline offset of the stable moment. Since the feature analysis time period is formed by expanding the complete physical response cycle, moments within two or more feature analysis time periods are recorded as overlapping moments. From the center moments of all feature analysis time periods where the overlapping moment is located, the center moment with the shortest time interval is selected. The actual baseline offset corresponding to the selected center moment is then used to correct the baseline offset of the overlapping moment.
[0093] Step 3: Detect the concentration of trace gases in the transformer oil.
[0094] The dynamic response curves of all electrical signals, after noise reduction and baseline offset correction, are synchronously input into a pre-trained BP neural network and extreme learning machine fusion detection model, which outputs the concentration values of trace gases in transformer oil. Specifically, in this embodiment... , The concentration value is determined to achieve accurate detection of trace gases in transformer oil under low-temperature conditions. Through the above processing, this application achieves [the following] under low-temperature conditions. Detection error ≤2ppm With a stable detection error of ≤0.2ppm and a long-term drift rate of ≤3%, the detection results can accurately reflect the internal insulation status and fault information of the transformer.
[0095] It should be noted that the training process of the BP neural network and Extreme Learning Machine (ELM) fusion detection model adopts a cascaded fusion architecture of "BP neural network feature extraction + Extreme Learning Machine (ELM) concentration regression". The input layer receives the dynamic response curves of all seed electrical signals after noise reduction and baseline offset correction. The BP neural network has two hidden layers: the first layer has 12 neurons and the second layer has 8 neurons. The Sigmoid activation function is used to purify the concentration-sensitive features and output a 6-dimensional refined feature vector. The ELM has one randomly mapped hidden layer with 15 neurons. The input weights and biases are randomly initialized in the interval [-1,1]. The output layer uses the least squares method to analytically solve for the output weights. During training, the mean squared error is used as the loss function to quantify the deviation between the predicted concentration of the BP neural network and ELM fusion detection model and the actual concentration of the sample. The BP neural network uses the Adam optimizer to iteratively update the parameters until the verification loss decreases by less than 5 times consecutively. The model converges quickly, and ELM does not require iterative optimization; the output weights are directly solved using the least squares method. After training, the dynamic response curves of all electrical signals, after noise reduction and baseline offset correction, are input into the BP neural network and extreme learning machine fusion detection model. The model then sequentially performs feature extraction via BP neural network and concentration regression via ELM to output high-precision concentration detection results. The BP neural network and extreme learning machine fusion detection model uses BP neural network to extract complex interference features and extreme learning machine to achieve rapid and high-precision concentration inversion, thus enabling accurate detection of trace gas concentrations in transformer oil. The detailed model training process is well-known to those skilled in the art and will not be elaborated upon in this application.
[0096] It should be added that: In this application, when calculating the ratio, if there is a case where the denominator is 0, the denominator is first mapped to a positive number before subsequent calculations are performed. There are many ways to map data to a positive number, and implementers can choose existing feasible methods according to the actual situation. In this embodiment, the purpose of mapping the data to a positive number is achieved by calculating the sum of the data and a preset constant greater than 0. The value of the preset constant greater than 0 is preset by humans, and implementers can set it according to the actual situation. This application does not impose any special restrictions. In this embodiment, the value of the preset constant greater than 0 is 0.01.
[0097] In summary, this application, by analyzing the temporal fluctuation characteristics of single environmental data and dividing time intervals, can accurately capture the phased fluctuation patterns of single environmental data during concentration detection; by comparing the number of data in each time interval with that in a preset time period, it reflects the continuous impact range of the fluctuation changes of single environmental data; by calculating the dispersion in each time interval, it quantifies the intensity of environmental data fluctuations; and by combining the proportion of data and the dispersion to obtain the phased interference degree, it can comprehensively evaluate the significance of the impact of single environmental data on the sensor in different time intervals, achieve a phased and refined assessment of environmental interference analysis, and provide a single-factor reference for subsequently determining the comprehensive interference level at each moment.
[0098] Furthermore, by measuring the outlier degree of various environmental data at different times, it is possible to accurately identify the moments when there are abnormal environmental disturbances. This can then be combined with the stage interference degree to obtain the comprehensive interference significance, taking into account both the stage fluctuation characteristics of environmental data and the degree of deviation of single-point anomalies, and achieving a comprehensive quantitative assessment of multiple environmental interference factors. In addition, the significant interference moments can be screened to determine the feature analysis time period, which can accurately locate the time period when the sensor is most severely affected by comprehensive interference.
[0099] Furthermore, within each characteristic analysis time period, by measuring the characteristic deviation of the dynamic response curves of various electrical signals relative to the reference response curve, a reliable basis is provided for the subsequent dynamic adjustment of the baseline offset, so that the baseline correction intensity is dynamically matched with the actual interference level, improving the error compensation accuracy, enhancing the authenticity and stability of the dynamic response curve, and thus improving the detection accuracy of trace gas concentration in transformer oil under low temperature conditions.
[0100] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0101] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from its essential characteristics. Therefore, the embodiments described above should be considered exemplary and non-limiting in all respects.
Claims
1. A high-precision detection method for trace gases in transformer oil based on error compensation, characterized in that, The method includes the following steps: Within a preset time period under a preset temperature environment, the dynamic response curves of various electrical signals generated by the sensor when detecting the concentration of trace gases in transformer oil are collected, as well as various environmental data at each time point. The environmental data includes ambient temperature, sensor surface temperature, sensor vibration amplitude, and substation electromagnetic interference intensity. By analyzing the fluctuations of individual environmental data, the data is divided into time intervals. The number of individual environmental data points within each time interval is compared with the number within the preset time period, and the intensity of fluctuations in the individual environmental data points within each time interval is also compared to obtain the stage interference degree of the individual environmental data points. The outlier degree of various environmental data points at each time point is measured and combined with the stage interference degree of all environmental data points to obtain the comprehensive interference significance at each time point, thereby filtering out the time points with significant interference. By identifying significant interference moments and their nearest time ranges, we obtain the characteristic analysis time periods. By analyzing the characteristic deviation of the dynamic response curves of various electrical signals relative to the baseline response curve within each characteristic analysis time period, we obtain the comprehensive influence coefficient of each characteristic analysis time period. This coefficient is then used to correct the theoretical baseline deviations obtained using the preset temperature drift model, resulting in the actual baseline deviations for each characteristic analysis time period. These deviations are then used to synchronously correct the dynamic response curves of various electrical signals in order to detect the concentration of trace gases in transformer oil. The process for obtaining the stage interference degree is as follows: Calculate the proportion of a single type of environmental data in each time interval among all single type of environmental data in the preset time period; Calculate the dispersion of the normalized values of a single environmental data type within each time interval; The stage interference degree is obtained by combining the quantity proportion and the dispersion across all time intervals. The process for obtaining the comprehensive interference significance is as follows: The product of the normalized value of the stage interference of various environmental data and the measure of the outlier degree of various environmental data at each time point is used as the significance of the influence of various environmental data at each time point. The comprehensive interference significance is the sum of the influence significance of all environmental data at each time point.
2. The high-precision detection method for trace gases in transformer oil based on error compensation as described in claim 1, characterized in that, The calculation process for the stage interference degree is as follows: The stage characteristic values of a single type of environmental data in each time interval are obtained by the proportion of the quantity of the single type of environmental data in each time interval and the dispersion, and the stage characteristic values are positively correlated with the proportion of the quantity and the dispersion, respectively. The stage interference degree is the sum of the stage characteristic values of a single type of environmental data across all time intervals.
3. The high-precision detection method for trace gases in transformer oil based on error compensation as described in claim 1, characterized in that, The moment when the interference becomes significant is the moment when the overall interference significance is greater than a preset threshold.
4. The high-precision detection method for trace gases in transformer oil based on error compensation as described in claim 1, characterized in that, The process for obtaining the comprehensive influence coefficient is as follows: For a single feature analysis time period, the preset curve feature values of the dynamic response curve and the reference response curve of various electrical signals are extracted respectively, and the difference of all curve feature values between the dynamic response curve of various electrical signals and the reference response curve is comprehensively measured. The comprehensive influence coefficient for a single feature analysis time period is obtained by using the difference between the dynamic response curves of all types of electrical signals and their reference response curves.
5. The high-precision detection method for trace gases in transformer oil based on error compensation as described in claim 4, characterized in that, The comprehensive influence coefficient of the single feature analysis time period is the weighted sum of the normalized values of the differences between the dynamic response curves of all electrical signals and their reference response curves.
6. The high-precision detection method for trace gases in transformer oil based on error compensation as described in claim 1, characterized in that, The actual baseline offset is the product of the comprehensive influence coefficient and the theoretical baseline offset.
7. The high-precision detection method for trace gases in transformer oil based on error compensation as described in claim 1, characterized in that, In the process of synchronously correcting the dynamic response curves of various electrical signals, the time not within any feature analysis time period is recorded as a stable time. The theoretical baseline offset is obtained using a preset temperature drift model. The theoretical baseline offset is used as the actual baseline offset to correct the baseline offset of the stable time. The time within two or more feature analysis time periods is recorded as a coincident time. From the center time of all feature analysis time periods in which the coincident time is located, the center time with the shortest time interval between it and the coincident time is selected. The actual baseline offset corresponding to the selected center time is used to correct the baseline offset of the coincident time.