Methods, apparatus, equipment and storage media for determining the concentration of decomposition products of insulating gases

Abstract

This invention relates to the field of gas decomposition product analysis technology, and particularly to a method, apparatus, device, and storage medium for determining the concentration of insulating gas decomposition products. The method first acquires multiple gas concentration data queues; then, it segments and aggregates these data queues into multiple segmented datasets, and constructs a frequency domain matrix from multiple frequency domain feature vectors extracted from each segmented dataset; next, it performs convolution and pooling operations on each frequency domain matrix to obtain a feature matrix; finally, it sequentially extracts matrices from the multiple feature matrices and inputs them into a gas concentration prediction model. When the fluctuation of the model's output is less than a threshold, the concentration of various gases is determined based on the model's output. This invention determines the concentration of insulating gas decomposition products by repeatedly inputting the feature matrix and when the output value is stable. Therefore, it can obtain stable sensor values ​​in a timely manner and eliminates the influence of different gases on the sensor, resulting in more accurate concentration results.

Description

Technical Field

[0001] This invention relates to the field of gas decomposition product analysis technology, and in particular to a method, apparatus, equipment and storage medium for determining the concentration of insulating gas decomposition products. Background Technology

[0002] SF6 gas is a colorless, odorless, non-toxic, and non-flammable inert gas. Due to its high electronegativity, it has unique insulating properties and efficient arc-extinguishing performance, making it one of the most important insulating media in the power industry and widely used in gas-insulated switchgear (GIS).

[0003] In GIS electrical equipment, the insulation performance of SF6 gas / mixed gas directly affects the safe operation of the equipment. By detecting the concentration of SF6 gas / mixed gas / mixed gas, potential insulation faults can be detected in a timely manner, thereby preventing safety accidents caused by insulation problems. The density, pressure, purity within the gas chamber, trace moisture content, and decomposition products of SF6 gas / mixed gas are important indicators for judging the operating status of GIS electrical equipment. By detecting these indicators, it is possible to determine whether there are leaks inside the GIS electrical equipment, whether the equipment is properly sealed, etc., thus enabling the timely detection and handling of potential safety hazards.

[0004] Methods for detecting insulating gases and decomposition products in high-voltage electrical equipment include electrochemical methods, chromatography, and spectrometry. Currently, the mainstream approach for detecting insulating gases and decomposition products is to use multiple sensors to detect different gas components. The accuracy of the detection results returned by the sensors is mainly affected by two factors: gas diffusion and the influence of other gas components. The effect of gas diffusion means that after a change in gas composition, a period of diffusion is required before the sensor returns a relatively accurate detection value. The influence of other gas components means that different gas components can affect the detection value. For example, in an SF6 / N2 mixed gas environment, the presence of N2 may interfere with the detection of certain decomposition products.

[0005] Currently, some technologies correct the detection results of insulating gases and decomposition products by using noise removal algorithms or data comparison reference methods. The advantage of these methods is that the algorithms are relatively simple and easy to implement, but the disadvantage is that they have very limited effect on improving the accuracy and timeliness of the detection results of insulating gases and decomposition products.

[0006] Therefore, it is necessary to develop a method for determining the concentration of insulating gas decomposition products. Summary of the Invention

[0007] The present invention provides a method, apparatus, device, and storage medium for determining the concentration of insulating gas decomposition products, which solves the problem of inaccurate detection results of insulating gas and decomposition products in the prior art due to the influence of other gas components.

[0008] In a first aspect, embodiments of the present invention provide a method for determining the concentration of insulating gas decomposition products, comprising:

[0009] Multiple gas concentration data queues are acquired, where each gas concentration data queue characterizes the properties of gas concentration changes over time;

[0010] The multiple gas concentration data queues are divided and aggregated into multiple segmented datasets, and multiple frequency domain feature vectors extracted from each segmented dataset are used to construct a frequency domain matrix. Each segmented dataset includes data segments extracted from multiple gas concentration data queues. The multiple data segments in the segmented dataset correspond to the same segmented time period, and the frequency domain feature vectors are extracted based on the data segments.

[0011] Perform convolution and pooling operations on each frequency domain matrix to obtain the feature matrix;

[0012] Matrix is ​​extracted sequentially from multiple feature matrices and input into the gas concentration prediction model. When the fluctuation of the model output is less than a threshold, the concentration of various gases is determined based on the model output.

[0013] In one possible implementation, the step of segmenting and aggregating the multiple gas concentration data queues into multiple segmented datasets, and constructing a frequency domain matrix from multiple frequency domain feature vectors extracted from each segmented dataset, includes:

[0014] Obtain the basic cycle duration;

[0015] The segment length is determined based on the basic cycle duration, wherein the segment length is an integer multiple of the basic cycle duration;

[0016] For each segmented time period, perform the following steps:

[0017] Data segments are extracted from the multiple gas concentration data queues according to the segmentation time period to construct a segmented dataset;

[0018] According to the preset gas concentration order, data segments are extracted from the segmented dataset as data segments to be processed;

[0019] Multiple frequency domain features are extracted from the data segment to be processed according to the first formula, and the extracted multiple frequency domain features are constructed into a frequency domain feature vector, wherein the first formula is:

[0020]

[0021] In the formula, GF(k) is the kth frequency domain feature, qnMax is the total number of data in the data segment to be processed, QCD(qn) is the qnth data in the data segment to be processed, e is the natural constant, j is the imaginary unit, π is pi, ω0 is the frequency corresponding to the basic period duration, and tN is the number of gas concentration data acquired within the basic period duration.

[0022] Add the frequency domain feature vector to the frequency domain matrix;

[0023] If the traversal of the segmented dataset is not completed, the process jumps to the step of extracting data segments from the segmented dataset as data segments to be processed according to a preset gas order.

[0024] In one possible implementation, the convolution and pooling operations on each frequency domain matrix to obtain the feature matrix include:

[0025] For each frequency domain matrix, perform the following steps:

[0026] Obtain the convolution operator;

[0027] Based on the position index, extract data blocks of the same type as the convolution operator from the frequency domain matrix as data blocks to be processed;

[0028] The convolution operator is performed with the data block to be processed to obtain the dot product result.

[0029] The dot product result is added to the convolution matrix according to the position index;

[0030] If the traversal of the frequency domain matrix is ​​not completed, the position index is offset according to the preset step length, and the process jumps to the step of extracting a data block of the same type as the convolution operator from the frequency domain matrix according to the position index as a data block to be processed.

[0031] Otherwise, max pooling is performed on the convolution matrix to obtain the feature matrix.

[0032] In one possible implementation, the step of sequentially extracting matrices from multiple feature matrices and inputting them into a gas concentration prediction model, and determining the concentrations of multiple gases based on the model's output when the fluctuation of the model's output is less than a threshold, includes:

[0033] Extract matrices sequentially from multiple feature matrices to serve as the input matrix;

[0034] The input matrix is ​​fed into the gas concentration prediction model to obtain multiple second outputs;

[0035] Based on the second formula, the plurality of second outputs, and the plurality of first outputs, the output deviation rate is determined, wherein the plurality of first outputs are the previous outputs of the gas concentration prediction model, and the second formula is:

[0036]

[0037] In the formula, Bias is the output deviation rate. For the cnth second output, cn represents the first output of the cnth generation, and cnMax is the total number of output nodes of the gas concentration prediction model.

[0038] Add the output deviation rate to the deviation rate queue;

[0039] If the values ​​of the last few bits of the deviation rate queue are all less than the deviation rate threshold, then the multiple second outputs are used as the concentrations of the various gases.

[0040] Otherwise, proceed to the step of sequentially extracting matrices from multiple feature matrices as the input matrix.

[0041] In one possible implementation, the gas concentration prediction model is constructed based on multiple historical feature matrices, and the construction process of the gas concentration prediction model includes:

[0042] Obtain the prediction base model, multiple historical feature matrices, and multiple historical gas concentration arrays. Each historical feature matrix corresponds to a historical gas concentration array, and the time node corresponding to the historical gas concentration array is later than the time node corresponding to the historical feature matrix.

[0043] Multiple coefficient arrays are randomly generated, and the multiple coefficient arrays are substituted into the prediction base model to obtain multiple intermediate models;

[0044] Matrix is ​​extracted from the multiple historical feature matrices according to the order of time nodes, and used as the historical feature matrix to be input;

[0045] The historical feature matrix to be input is input into the multiple intermediate models respectively to obtain multiple output arrays, wherein each output array corresponds to one intermediate model;

[0046] Based on the multiple output arrays and the target historical gas concentration array, multiple model biases are determined, wherein the target historical gas concentration array is the historical gas concentration array corresponding to the historical feature matrix to be input;

[0047] Add the multiple model deviations to their respective deviation queues;

[0048] If there is a queue in which the last few characters are less than the deviation threshold among multiple deviation queues, then the intermediate model corresponding to the queue with the last few characters less than the deviation threshold shall be used as the gas concentration prediction model.

[0049] Otherwise, the coefficients of the multiple intermediate models are adjusted according to the multiple deviation queues, and the process jumps to the step of retrieving a matrix from the multiple historical feature matrices according to the order of time nodes, as the historical feature matrix to be input.

[0050] In one possible implementation, the prediction base model is:

[0051]

[0052] In the formula, QC cn To predict the cn-th output of the base model, where mMax is the total number of rows in the intermediate nodes, nMax is the total number of columns in the intermediate nodes, and wo cn,a wh is the a-th coefficient of the c-th output node. m,n,a h is the a-th coefficient of the node in the m-th row and n-th column. a,n-1 This is the output of the middle node in row a-(n-1)th column, where n k Who is the column number of the feed node. m,n,cn x is the cn-th feed coefficient of the feed-in column node in the m-th row and n-th column. b For the b-th input variable, wi m,1,b Let b be the coefficient of the middle node in the first column of the m-th row, bmax be the total number of input variables, e be the natural constant, and C be the bias coefficient.

[0053] In one possible implementation, adjusting the coefficients of the plurality of intermediate models according to the plurality of deviation queues includes:

[0054] For each intermediate model, perform the following steps:

[0055] Determine the historical best coefficient array based on the minimum value in the deviation queue;

[0056] The global optimal coefficient array is determined based on the minimum value among the last values ​​of the multiple deviation queues;

[0057] The coefficients of the intermediate model are adjusted according to the third formula, which is:

[0058]

[0059] In the formula, For the a-th coefficient in the intermediate model before adjustment, w a,opta w is the a-th coefficient in the globally optimal coefficient array. a,opth This is the a-th coefficient in the historical best coefficient array. The globally optimal distance. The historical best distance, k1 is the first coefficient, k2 is the second coefficient, and k a k is the first substitution coefficient. h The second substitution coefficient, denoted as the a-th coefficient of the adjusted intermediate model, and distb as the basic perturbation.

[0060] Secondly, embodiments of the present invention provide an insulating gas decomposition product concentration device for implementing the insulating gas decomposition product concentration determination method as described in the first aspect or any possible implementation thereof, the insulating gas decomposition product concentration device comprising:

[0061] The gas concentration data acquisition module is used to acquire multiple gas concentration data queues, where each gas concentration data queue represents the characteristics of gas concentration changes over time.

[0062] The frequency domain analysis module is used to divide the queue time period into multiple segment time periods and construct a frequency domain matrix from multiple frequency domain feature vectors extracted from each segment dataset. Each segment dataset includes data segments extracted from multiple gas concentration data queues. Multiple data segments in the segment dataset correspond to the same segment time period, and the frequency domain feature vector is extracted based on the data segments.

[0063] The feature extraction module is used to perform convolution and pooling operations on each frequency domain matrix to obtain the feature matrix;

[0064] as well as,

[0065] The gas concentration determination module is used to sequentially extract matrices from multiple feature matrices and input them into the gas concentration prediction model. When the fluctuation of the model's output is less than a threshold, the concentration of various gases is determined based on the model's output.

[0066] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor executes the computer program to implement the steps of the method as described in the first aspect or any possible implementation of the first aspect.

[0067] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described in the first aspect or any possible implementation thereof.

[0068] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows:

[0069] This invention discloses a method for determining the concentration of decomposition products of insulating gases. The method first acquires multiple gas concentration data queues, each representing the characteristics of gas concentration changes over time. Then, the multiple gas concentration data queues are segmented and aggregated into multiple segmented datasets. Multiple frequency domain feature vectors extracted from each segmented dataset are used to construct a frequency domain matrix. Each segmented dataset includes data segments extracted from the multiple gas concentration data queues, with the data segments corresponding to the same time segment. The frequency domain feature vectors are extracted based on the data segments. Next, convolution and pooling operations are performed on each frequency domain matrix to obtain a feature matrix. Finally, matrices are sequentially extracted from the multiple feature matrices and input into a gas concentration prediction model. When the fluctuation of the model's output is less than a threshold, the concentrations of various gases are determined based on the model's output. This invention extracts frequency domain features from the data obtained by segmenting gas concentration data queues and constructs a frequency domain matrix, thus frequency-domainizing the time-series information, reducing data dimensionality, and including more factor information in the frequency domain matrix, providing data assurance for determining decomposition product concentrations. This invention reduces data dimensionality through convolution and pooling, reducing computational load and preventing overfitting, resulting in more accurate data. The present invention determines the concentration of insulating gas decomposition products by inputting the feature matrix multiple times and when the output value is stable. Therefore, the stable value of the sensor can be obtained in a timely manner, and the influence of different gases on the sensor is eliminated, resulting in more accurate concentration results. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0071] Figure 1 This is a flowchart of the method for determining the concentration of insulating gas decomposition products provided in the embodiments of the present invention;

[0072] Figure 2 This is a schematic diagram illustrating the construction process of the gas concentration prediction model provided in the embodiments of the present invention;

[0073] Figure 3 This is a functional block diagram of the insulating gas decomposition product concentration device provided in the embodiments of the present invention;

[0074] Figure 4 This is a functional block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0075] In the following description, specific details such as particular system structures and techniques are set forth for illustrative purposes and not for limitation, so as to provide a thorough understanding of embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0076] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.

[0077] The embodiments of the present invention will be described in detail below. This example is implemented based on the technical solution of the present invention, and provides detailed implementation methods and specific operation processes. However, the protection scope of the present invention is not limited to the following embodiments.

[0078] Figure 1 A flowchart illustrating the method for determining the concentration of insulating gas decomposition products provided in an embodiment of the present invention.

[0079] like Figure 1 As shown, a flowchart illustrating the implementation of the method for determining the concentration of insulating gas decomposition products provided by an embodiment of the present invention is illustrated below:

[0080] In step 101, multiple gas concentration data queues are acquired, wherein the gas concentration data queues characterize the properties of gas concentration changes over time.

[0081] In step 102, the multiple gas concentration data queues are divided and aggregated into multiple segmented datasets, and multiple frequency domain feature vectors extracted from each segmented dataset are used to construct a frequency domain matrix. Each segmented dataset includes data segments extracted from multiple gas concentration data queues, and the multiple data segments in the segmented dataset correspond to the same segmented time period. The frequency domain feature vectors are extracted based on the data segments.

[0082] In some implementations, the step of segmenting and aggregating the multiple gas concentration data queues into multiple segmented datasets, and constructing a frequency domain matrix from multiple frequency domain feature vectors extracted from each segmented dataset, includes:

[0083] Obtain the basic cycle duration;

[0084] The segment length is determined based on the basic cycle duration, wherein the segment length is an integer multiple of the basic cycle duration;

[0085] For each segmented time period, perform the following steps:

[0086] Data segments are extracted from the multiple gas concentration data queues according to the segmentation time period to construct a segmented dataset;

[0087] According to the preset gas concentration order, data segments are extracted from the segmented dataset as data segments to be processed;

[0088] Multiple frequency domain features are extracted from the data segment to be processed according to the first formula, and the extracted multiple frequency domain features are constructed into a frequency domain feature vector, wherein the first formula is:

[0089]

[0090] In the formula, GF(k) is the kth frequency domain feature, qnMax is the total number of data in the data segment to be processed, QCD(qn) is the qnth data in the data segment to be processed, e is the natural constant, j is the imaginary unit, π is pi, ω0 is the frequency corresponding to the basic period duration, and tN is the number of gas concentration data acquired within the basic period duration.

[0091] Add the frequency domain feature vector to the frequency domain matrix;

[0092] If the traversal of the segmented dataset is not completed, the process jumps to the step of extracting data segments from the segmented dataset as data segments to be processed according to a preset gas order.

[0093] For example, as mentioned above, in current technologies, when the gas concentration changes, it takes a certain amount of time for the sensor's detection value to stabilize. Furthermore, due to the presence of multiple gases, the mutual interference between the gases can affect the detection value returned by the sensor.

[0094] The present invention aims to determine the stable value when the sensor detection value is stable from the early data stream returned by the sensor, and to eliminate as much interference as possible from other gases on the sensor's returned data.

[0095] The gas concentration data queue of this invention is obtained based on the return value of the sensor. Generally speaking, one gas concentration data queue corresponds to one sensor, that is, it corresponds to the concentration value of one gas.

[0096] To facilitate data stream analysis, this invention divides each data stream (gas concentration data queue) into multiple segments, extracts frequency domain features from data corresponding to the same time period, and constructs a frequency domain matrix. Thus, each time period corresponds to a frequency domain matrix. Since the matrix combines time-domain features with various gas concentration data features, this invention's concentration determination method is a gas concentration determination method based on time series and factor analysis.

[0097] In terms of frequency domain matrix construction, this invention first obtains the basic period duration, which is generally determined by the duration of the insulating gas during anomalies. For example, an integer multiple of the typical discharge duration of GIS equipment is used as the basic period. Based on this basic period duration, the segmented time period length is determined. Note that the segmented time period length is an integer multiple of the basic period duration. Then, data segments are extracted from each gas concentration data queue according to the segmented time period, and the resulting multiple data segments are constructed into a segmented dataset.

[0098] Then, according to the preset gas concentration order, data segments are extracted from the segmented dataset for frequency domain feature extraction. In one application scenario, the first formula is used, which is:

[0099]

[0100] In the formula, GF(k) is the kth frequency domain feature, qnMax is the total number of data in the data segment to be processed, QCD(qn) is the qnth data in the data segment to be processed, e is the natural constant, j is the imaginary unit, π is pi, ω0 is the frequency corresponding to the basic period duration, and tN is the number of gas concentration data acquired within the basic period duration.

[0101] The frequency domain features obtained through the above formula are used to construct frequency domain feature vectors and added to the frequency domain matrix. By repeating this process, data segments are extracted from the segmented dataset, frequency domain vectors are extracted, and added to the frequency domain matrix, thus completing the construction of the frequency domain matrix.

[0102] In step 103, convolution and pooling operations are performed on each frequency domain matrix to obtain the feature matrix.

[0103] In some implementations, the step of performing convolution and pooling operations on each frequency domain matrix to obtain a feature matrix includes:

[0104] For each frequency domain matrix, perform the following steps:

[0105] Obtain the convolution operator;

[0106] Based on the position index, extract data blocks of the same type as the convolution operator from the frequency domain matrix as data blocks to be processed;

[0107] The convolution operator is performed with the data block to be processed to obtain the dot product result.

[0108] The dot product result is added to the convolution matrix according to the position index;

[0109] If the traversal of the frequency domain matrix is ​​not completed, the position index is offset according to the preset step length, and the process jumps to the step of extracting a data block of the same type as the convolution operator from the frequency domain matrix according to the position index as a data block to be processed.

[0110] Otherwise, max pooling is performed on the convolution matrix to obtain the feature matrix.

[0111] For example, if the concentration of gaseous decomposition products is determined based on the frequency domain matrix obtained in the aforementioned steps, the computational complexity will be excessive due to the large dimension of the frequency domain matrix. Furthermore, the frequency vectors in the frequency domain matrix contain some low-value and continuous elements, which have a relatively small impact on the determination of decomposition product concentrations. This embodiment of the invention reduces the dimensionality of the frequency domain matrix obtained in the aforementioned steps through convolution and pooling operations.

[0112] First, we obtain the convolution operator, for example, a difference operator is:

[0113]

[0114] Based on the obtained difference operator and position index, extract the data block of the same type as the convolution operator from the frequency domain matrix, and perform a dot product operation between this data block and the data block to be processed to obtain the dot product result;

[0115] Next, the dot product result is added to the convolution matrix according to the position index. Then, the position index is adjusted, and the dot product result calculation steps are repeated to complete the convolution operation of the entire matrix, resulting in the convolution matrix.

[0116] Max pooling of the convolution matrix yields the feature matrix.

[0117] Max pooling is a downsampling operation. Its main purpose is to reduce the dimensionality of the input features, thereby reducing the computational cost and the number of parameters, while extracting the most salient feature information.

[0118] Suppose we have a feature map of size *n* (where *n* and *n* represent the number of rows and columns of the feature map, respectively). We define a pooling window with size *n* (usually *n* or *n*) and a stride. The pooling window starts from the top left corner of the feature map and slides across it according to the stride. For each sub-region covered by the pooling window, the maximum value in that sub-region is selected as the output value.

[0119] Max pooling can reduce computational cost, especially when dealing with large-scale datasets where computation is extremely high. Max pooling reduces data size, thus reducing the computational burden of subsequent layers. Max pooling can prevent overfitting: Overfitting occurs when a model performs well on training data but poorly on test data. Max pooling can be seen as an abstraction of the data, retaining only the most salient features while discarding some local details, resulting in better generalization ability and helping to prevent overfitting. Max pooling can extract key features: It effectively extracts the main features from the data. Because max pooling selects the maximum value within the pooling window, these maximum values ​​typically represent the most salient features in that local region, which is beneficial for subsequent tasks such as classification.

[0120] In step 104, matrices are sequentially extracted from multiple feature matrices and input into the gas concentration prediction model. When the fluctuation of the model's output is less than a threshold, the concentrations of various gases are determined based on the model's output.

[0121] In some implementations, the step of sequentially extracting matrices from multiple feature matrices and inputting them into a gas concentration prediction model, and determining the concentrations of multiple gases based on the model's output when the fluctuation of the model's output is less than a threshold, includes:

[0122] Extract matrices sequentially from multiple feature matrices to serve as the input matrix;

[0123] The input matrix is ​​fed into the gas concentration prediction model to obtain multiple second outputs;

[0124] Based on the second formula, the plurality of second outputs, and the plurality of first outputs, the output deviation rate is determined, wherein the plurality of first outputs are the previous outputs of the gas concentration prediction model, and the second formula is:

[0125]

[0126] In the formula, Bias is the output deviation rate. For the cnth second output, cn represents the first output of the cnth generation, and cnMax is the total number of output nodes of the gas concentration prediction model.

[0127] Add the output deviation rate to the deviation rate queue;

[0128] If the values ​​of the last few bits of the deviation rate queue are all less than the deviation rate threshold, then the multiple second outputs are used as the concentrations of the various gases.

[0129] Otherwise, proceed to the step of sequentially extracting matrices from multiple feature matrices as the input matrix.

[0130] In some implementations, the gas concentration prediction model is constructed based on multiple historical feature matrices, and the construction process of the gas concentration prediction model includes:

[0131] Obtain the prediction base model, multiple historical feature matrices, and multiple historical gas concentration arrays. Each historical feature matrix corresponds to a historical gas concentration array, and the time node corresponding to the historical gas concentration array is later than the time node corresponding to the historical feature matrix.

[0132] Multiple coefficient arrays are randomly generated, and the multiple coefficient arrays are substituted into the prediction base model to obtain multiple intermediate models;

[0133] Matrix is ​​extracted from the multiple historical feature matrices according to the order of time nodes, and used as the historical feature matrix to be input;

[0134] The historical feature matrix to be input is input into the multiple intermediate models respectively to obtain multiple output arrays, wherein each output array corresponds to one intermediate model;

[0135] Based on the multiple output arrays and the target historical gas concentration array, multiple model biases are determined, wherein the target historical gas concentration array is the historical gas concentration array corresponding to the historical feature matrix to be input;

[0136] Add the multiple model deviations to their respective deviation queues;

[0137] If there is a queue in which the last few characters are less than the deviation threshold among multiple deviation queues, then the intermediate model corresponding to the queue with the last few characters less than the deviation threshold shall be used as the gas concentration prediction model.

[0138] Otherwise, the coefficients of the multiple intermediate models are adjusted according to the multiple deviation queues, and the process jumps to the step of retrieving a matrix from the multiple historical feature matrices according to the order of time nodes, as the historical feature matrix to be input.

[0139] In some implementations, the prediction base model is:

[0140]

[0141] In the formula, QC cn To predict the cn-th output of the base model, where mMax is the total number of rows in the intermediate nodes, nMax is the total number of columns in the intermediate nodes, and wo cn,a wh is the a-th coefficient of the c-th output node. m,n,a h is the a-th coefficient of the node in the m-th row and n-th column. a，n-1 This is the output of the middle node in row a-(n-1)th column, where n k Who is the column number of the feed node.m,n,cn x is the cn-th feed coefficient of the feed-in column node in the m-th row and n-th column. b For the b-th input variable, wi m,1,b Let b be the coefficient of the middle node in the first column of the m-th row, bmax be the total number of input variables, e be the natural constant, and C be the bias coefficient.

[0142] In some implementations, adjusting the coefficients of the plurality of intermediate models according to the plurality of deviation queues includes:

[0143] For each intermediate model, perform the following steps:

[0144] Determine the historical best coefficient array based on the minimum value in the deviation queue;

[0145] The global optimal coefficient array is determined based on the minimum value among the last values ​​of the multiple deviation queues;

[0146] The coefficients of the intermediate model are adjusted according to the third formula, which is:

[0147]

[0148] In the formula, For the a-th coefficient in the intermediate model before adjustment, w a,opta w is the a-th coefficient in the globally optimal coefficient array. a,opth This is the a-th coefficient in the historical best coefficient array. The globally optimal distance. The historical best distance, k1 is the first coefficient, k2 is the second coefficient, and k a k is the first substitution coefficient. h The second substitution coefficient, denoted as the a-th coefficient of the adjusted intermediate model, and distb as the basic perturbation.

[0149] For example, in this invention, multiple feature matrices are sequentially input into a gas concentration prediction model. When the fluctuation of the model's output is less than a threshold, the concentrations of various gases are determined based on the model's output. This model has inputs corresponding to multiple elements of the feature matrices and multiple outputs corresponding to predicted gas concentration values.

[0150] like Figure 2 As shown in the figure, this diagram illustrates the principle of constructing the gas concentration prediction model provided in the embodiments of the present invention.

[0151] The construction process of the gas concentration prediction model provided in this embodiment of the invention first involves obtaining a prediction base model, multiple historical feature matrices 201, and multiple historical gas concentration arrays 202. In one application scenario, the prediction base model is:

[0152]

[0153] In the formula, QCcn is the cn-th output of the prediction base model, mMax is the total number of rows in the intermediate nodes, nMax is the total number of columns in the intermediate nodes, and wo cn,a wh is the a-th coefficient of the c-th output node. m,n,a h is the a-th coefficient of the node in the m-th row and n-th column. a,n-1 This is the output of the middle node in row a-(n-1)th column, where n k Who is the column number of the feed node. m,n,cn x is the cn-th feed coefficient of the feed-in column node in the m-th row and n-th column. b For the b-th input variable, wi m,1,b Let b be the coefficient of the middle node in the first column of the m-th row, bmax be the total number of input variables, e be the natural constant, and C be the bias coefficient.

[0154] We can see that there are quite a few coefficients in this model that need to be determined.

[0155] On the other hand, each historical feature matrix 201 has a one-to-one correspondence with each historical gas concentration array 202, and the time node corresponding to the historical gas concentration array 202 is later than the time node corresponding to the historical feature matrix 201.

[0156] Then, multiple coefficient arrays are randomly generated and substituted into the prediction base model to generate multiple intermediate models 203. Next, matrices are extracted from multiple historical feature matrices 201 according to the time node order and input into the multiple intermediate models 203 respectively to obtain multiple output arrays 204. Since each historical feature matrix 201 has a one-to-one correspondence with each historical gas concentration array 202, we can determine the model bias based on the output array 204 and the target historical gas concentration array. Note that the target historical gas concentration array is the historical gas concentration array corresponding to the input historical feature matrix. The obtained model bias is added to the corresponding bias queue 205. Note that each bias queue 205 corresponds to one intermediate model 203.

[0157] If there exists a queue with the last few values ​​less than the deviation threshold, the intermediate model 203 corresponding to this queue is used as the gas concentration prediction model, and the iteration ends. If multiple deviation queues do not meet the above conditions, the coefficients of multiple intermediate models 203 are adjusted according to the deviation queue 205, and the steps of retrieving matrices from multiple historical feature matrices 201 according to the time node order and inputting them into the intermediate model 203 are repeated. Through the above iterative process, the model can be constructed.

[0158] Regarding adjusting the coefficients of the intermediate model based on multiple deviation queues, this embodiment of the invention determines a historically optimal coefficient array based on the minimum value in each deviation queue; then, it determines a globally optimal coefficient array based on the minimum value at the end of each deviation queue; and finally, it adjusts the coefficients of the intermediate model according to the third formula:

[0159]

[0160] In the formula, For the a-th coefficient in the intermediate model before adjustment, w a,opta w is the a-th coefficient in the globally optimal coefficient array. a,opth This is the a-th coefficient in the historical best coefficient array. The globally optimal distance. The historical best distance, k1 is the first coefficient, k2 is the second coefficient, and k a k is the first substitution coefficient. h The second substitution coefficient, denoted as the a-th coefficient of the adjusted intermediate model, and distb as the basic perturbation.

[0161] This invention discloses a method for determining the concentration of decomposition products of insulating gases. First, it acquires multiple gas concentration data queues, each representing the characteristics of gas concentration changes over time. Then, it segments and aggregates these data queues into multiple segmented datasets. Multiple frequency domain feature vectors extracted from each segmented dataset are used to construct a frequency domain matrix. Each segmented dataset includes data segments extracted from the multiple gas concentration data queues, with these segments corresponding to the same time period. The frequency domain feature vectors are extracted based on these data segments. Next, convolution and pooling operations are performed on each frequency domain matrix to obtain a feature matrix. Finally, matrices are sequentially extracted from these feature matrices and input into a gas concentration prediction model. When the fluctuation of the model's output is less than a threshold, the concentrations of various gases are determined based on the model's output. This invention extracts frequency domain features from the data obtained by segmenting gas concentration data queues and constructs a frequency domain matrix, thus frequency-domainizing the time-series information, reducing data dimensionality, and including more factor information in the frequency domain matrix. This provides data assurance for determining decomposition product concentrations. Furthermore, this invention reduces data dimensionality through convolution and pooling, decreasing computational load and preventing overfitting, resulting in more accurate data. The present invention determines the concentration of insulating gas decomposition products by inputting the feature matrix multiple times and when the output value is stable. Therefore, the stable value of the sensor can be obtained in a timely manner, and the influence of different gases on the sensor is eliminated, resulting in more accurate concentration results.

[0162] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0163] The following are embodiments of the apparatus of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0164] Figure 3 This is a functional block diagram of the insulating gas decomposition product concentration device provided in an embodiment of the present invention, with reference to... Figure 3 The device for determining the concentration of insulating gas decomposition products includes: a gas concentration data acquisition module 301, a frequency domain analysis module 302, a feature extraction module 303, and a gas concentration determination module 304, wherein:

[0165] The gas concentration data acquisition module 301 is used to acquire multiple gas concentration data queues, wherein the gas concentration data queues characterize the gas concentration changes over time.

[0166] The frequency domain analysis module 302 is used to divide the queue period into multiple segmented periods and construct a frequency domain matrix from multiple frequency domain feature vectors extracted from each segmented dataset. Each segmented dataset includes data segments extracted from multiple gas concentration data queues. Multiple data segments in the segmented dataset correspond to the same segmented period, and the frequency domain feature vectors are extracted based on the data segments.

[0167] The feature extraction module 303 is used to perform convolution and pooling operations on each frequency domain matrix to obtain the feature matrix;

[0168] The gas concentration determination module 304 is used to extract matrix inputs from multiple feature matrices sequentially and input them into the gas concentration prediction model. When the fluctuation of the model output is less than a threshold, the concentration of multiple gases is determined based on the model output.

[0169] Figure 4 This is a functional block diagram of the electronic device provided in an embodiment of the present invention. For example... Figure 4 As shown, the electronic device 4 of this embodiment includes a processor 400 and a memory 401, wherein the memory 401 stores a computer program 402 that can run on the processor 400. When the processor 400 executes the computer program 402, it implements the steps of the above-described methods and embodiments for determining the concentration of insulating gas decomposition products, for example... Figure 1 Steps 101 to 104 are shown.

[0170] For example, the computer program 402 may be divided into one or more modules / units, which are stored in the memory 401 and executed by the processor 400 to complete the present invention.

[0171] The electronic device 4 can be a desktop computer, laptop, handheld computer, cloud server, or other computing device. The electronic device 4 may include, but is not limited to, a processor 400 and a memory 401. Those skilled in the art will understand that... Figure 4 This is merely an example of electronic device 4 and does not constitute a limitation on electronic device 4. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 4 may also include input / output devices, network access devices, buses, etc.

[0172] The processor 400 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0173] The memory 401 can be an internal storage unit of the electronic device 4, such as a hard disk or memory. The memory 401 can also be an external storage device of the electronic device 4, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 401 can include both internal and external storage units of the electronic device 4. The memory 401 is used to store the computer program 402 and other programs and data required by the electronic device 4. The memory 401 can also be used to temporarily store data that has been output or will be output.

[0174] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the aforementioned method embodiments, and will not be repeated here.

[0175] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0176] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0177] In the embodiments provided by this invention, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0178] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0179] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0180] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above-described embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various methods and apparatus embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0181] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method of determining a concentration of an insulating gas decomposition product, characterized by, The method comprises the following steps: obtaining a plurality of gas concentration data queues, wherein the gas concentration data queues represent the characteristics of the change of gas concentration over time; dividing and collecting the plurality of gas concentration data queues into a plurality of divided data sets, and constructing a plurality of frequency domain feature vectors extracted from each divided data set into a frequency domain matrix, wherein each divided data set comprises data segments extracted from the plurality of gas concentration data queues, and the plurality of data segments in the divided data set correspond to the same division period, and the frequency domain feature vector is extracted from the data segment; performing convolution and pooling operations on each frequency domain matrix to obtain a feature matrix; sequentially extracting matrix input into a gas concentration prediction model from a plurality of feature matrices, and determining the concentration of a plurality of gases according to the output of the model when the fluctuation of the output of the model is less than a threshold value; the gas concentration prediction model is constructed according to a plurality of historical feature matrices, and the construction process of the gas concentration prediction model comprises: obtaining a prediction base model, a plurality of historical feature matrices, and a plurality of historical gas concentration arrays, wherein each historical feature matrix corresponds to a historical gas concentration array, and the time node corresponding to the historical gas concentration array is later than the time node corresponding to the historical feature matrix; randomly generating a plurality of coefficient arrays, and substituting the plurality of coefficient arrays into the prediction base model to obtain a plurality of intermediate models; taking the matrix from the plurality of historical feature matrices according to the order of the time node as the input historical feature matrix; inputting the input historical feature matrix into the plurality of intermediate models respectively to obtain a plurality of output arrays, wherein each output array corresponds to an intermediate model; determining a plurality of model biases according to the plurality of output arrays and a target historical gas concentration array, wherein the target historical gas concentration array is the historical gas concentration array corresponding to the input historical feature matrix; adding the plurality of model biases to the corresponding bias queue respectively; if there are queues with the last few positions less than the bias threshold value in the plurality of bias queues, then the intermediate model corresponding to the queue with the last few positions less than the bias threshold value is taken as the gas concentration prediction model; otherwise, adjust the coefficients of the plurality of intermediate models according to the plurality of bias queues respectively, and jump to the step of taking the matrix from the plurality of historical feature matrices as the input historical feature matrix according to the order of the time node.

2. The method of claim 1, wherein The step of dividing and collecting the plurality of gas concentration data queues into a plurality of divided data sets, and constructing a plurality of frequency domain feature vectors extracted from each divided data set into a frequency domain matrix comprises: obtaining a basic cycle length; determining the division period length according to the basic cycle length, wherein the division period length is an integer multiple of the basic cycle length; for each division period, the following steps are performed respectively: constructing the data segment extracted from the plurality of gas concentration data queues according to the division period into a divided data set; extracting the data segment from the divided data set as a to-be-processed data segment according to a preset gas concentration order; extracting a plurality of frequency domain features of the to-be-processed data segment according to a first formula, and constructing the extracted plurality of frequency domain features into a frequency domain feature vector, wherein the first formula is: In the formula, GF(k) is the kth frequency domain feature, qnMax is the total number of data in the data segment to be processed, QCD(qn) is the qnth data of the data segment to be processed, e is a natural constant, j is an imaginary unit, π is a circular constant, ω0 is a frequency corresponding to a basic cycle length, and tN is the number of gas concentration data obtained within the basic cycle length. The frequency domain feature vector is added to a frequency domain matrix. If the traversal of the segmented data set is not completed, the step of extracting a data segment from the segmented data set as a data segment to be processed according to a preset gas sequence is performed.

3. The method of claim 1, wherein The convolution and pooling operation is performed on each frequency domain matrix to obtain a feature matrix, including: For each frequency domain matrix, the following steps are performed respectively: An convolution operator is obtained; A data block of the same type as the convolution operator is extracted from the frequency domain matrix according to the position index as a data block to be processed; The convolution operator and the data block to be processed are dot multiplied to obtain a dot product result; The dot product result is added to a convolution matrix according to the position index; If the traversal of the frequency domain matrix is not completed, the position index is offset by a preset step length, and the step of extracting a data block of the same type as the convolution operator from the frequency domain matrix according to the position index as a data block to be processed is performed; Otherwise, the convolution matrix is maximum-pooled to obtain a feature matrix.

4. The method of claim 1-3, wherein, The matrix is sequentially extracted from the plurality of feature matrices as an input matrix to the gas concentration prediction model, and when the fluctuation of the output of the model is less than a threshold value, the concentration of the plurality of gases is determined according to the output of the model, including: The matrix is sequentially extracted from the plurality of feature matrices as an input matrix; The input matrix is input to the gas concentration prediction model to obtain a plurality of second outputs; The output deviation rate is determined according to a second formula, the plurality of second outputs, and a plurality of first outputs, wherein the plurality of first outputs are a plurality of outputs of the gas concentration prediction model at a previous time, and the second formula is: In the formula, Bias is an output bias rate, is the cnth second output, is the cnth first output, and cnMax is the total number of gas concentration prediction model output nodes. The output deviation rate is added to a deviation rate queue; If the values of the last positions of the deviation rate queue are all less than a deviation rate threshold value, the plurality of second outputs are taken as the concentrations of the plurality of gases; Otherwise, the step of sequentially extracting a matrix from the plurality of feature matrices as an input matrix is performed.

5. The method of claim 1, wherein The prediction base model is: wherein QC cn is the cnth output of the prediction-based model, mMax is the total number of rows of the intermediate nodes, nMax is the total number of columns of the intermediate nodes, wo cn,a is the ath coefficient of the cnth output node, wh m,n,a is the ath coefficient of the mth row and nth column node, h a,n-1 is the output of the ath row and n-1th column intermediate node, n k is the number of columns of the feeding-in nodes, who m,n,cn is the cnth feeding-in coefficient of the mth row and nth column feeding-in column node, x b is the bth input variable, wi m,1,b is the bth coefficient of the mth row and first column intermediate node, bmax is the total number of input variables, e is the natural constant, and C is the bias coefficient.

6. The method of claim 1, wherein The coefficients of the plurality of intermediate models are adjusted according to the plurality of deviation queues, including: For each intermediate model, the following steps are performed respectively: A historical optimal coefficient array is determined according to the minimum value in the deviation queue; A global optimal coefficient array is determined according to the minimum value in the last position of the plurality of deviation queues; The coefficients of the intermediate model are adjusted according to a third formula, wherein the third formula is: wherein, wa is the a-th coefficient of the intermediate model before adjustment a,opta wa is the a-th coefficient of the global optimal coefficient array a,opth wa is the a-th coefficient of the historical optimal coefficient array dista is the global optimal distance dista is the historical optimal distance, k1 is the first coefficient, k2 is the second coefficient, k a k1 is the first coefficient h k2 is the second coefficient wa is the a-th coefficient of the intermediate model after adjustment, distb is the basic perturbation 7. An insulation gas decomposate concentration device characterized by comprising: The device for determining the concentration of the decomposition product of the insulating gas includes: A gas concentration data acquisition module is configured to acquire a plurality of gas concentration data queues, wherein the gas concentration data queues represent the characteristics of the change of the gas concentration over time. The frequency domain analysis module is configured to divide the queue period into multiple division periods, and construct multiple frequency domain feature vectors extracted according to each division data set into a frequency domain matrix, wherein each division data set comprises data segments extracted from the multiple gas concentration data queues, the multiple data segments in the division data set correspond to the same division period, and the frequency domain feature vector is extracted according to the data segment. The feature extraction module is configured to perform convolution and pooling operations on each frequency domain matrix to obtain a feature matrix. In addition, the gas concentration determination module is configured to sequentially extract matrix inputs into a gas concentration prediction model from the multiple feature matrices, and determine the concentrations of multiple gases according to the output of the model when the fluctuation of the output of the model is less than a threshold.

8. An electronic device comprising a memory and a processor, said memory having stored therein a computer program operable on said processor, characterized in that, The processor executes the computer program to realize the steps of the method of any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to realize the steps of the method of any one of claims 1-6.

Images