Transformer risk assessment method and system based on learning numerical pulse neural membrane

Abstract

The present invention discloses a transformer risk assessment method and system based on a learning numerical pulse neural membrane, the method comprising: obtaining basic data of power transformers and constructing a power transformer data set, the basic data of power transformers referring to oil chromatography data; based on the basic data of power transformers, analyzing the generation mechanism of dissolved gas in transformer insulating oil, and analyzing the relationship between dissolved gas in transformer insulating oil and risk level, and obtaining transformer status; constructing a learning numerical pulse neural membrane calculation model, and adding a dimension-raising structure and a weight adaptive adjustment strategy to the learning numerical pulse neural membrane calculation model; according to the power transformer data set, performing transformer risk assessment based on the learning numerical pulse neural membrane calculation model, and obtaining an assessment result, i.e., transformer status. The present invention can cope with complex fault modes and capture these deep-level features, and has a high assessment accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of power system and transformer risk assessment, and in particular to a transformer risk assessment method and system based on a learning numerical pulse neural membrane. Background Art

[0002] As the core equipment in the power system, the stable operation of the transformer is crucial to the reliability and economy of the power grid. However, as the equipment ages, the transformer may pose huge safety risks due to aging, overload or external factors.

[0003] Common transformer risk assessment mainly uses a combination of oil chromatography and proportional analysis to determine the transformer risk type. However, this type of method mainly relies on expert experience and statistical models. Although simple, it is difficult to deal with complex failure modes and capture these deep-level characteristics.

[0004] In view of this, this application is hereby filed. Summary of the invention

[0005] The technical problem to be solved by the present invention is that in the prior art, transformer risk assessment mainly adopts a combination of oil chromatography analysis and proportional analysis methods to determine the transformer risk type. Such methods mainly rely on expert experience and statistical models, and are difficult to deal with complex fault modes and capture these deep-level features, and the assessment accuracy is not high. The purpose of the present invention is to provide a transformer risk assessment method and system based on a learning numerical pulse neural membrane. First, the gas generation mechanism and operating characteristics in the oil are analyzed, and then the relationship between the dissolved gas in the transformer oil and the risk level is analyzed. Then, a learning numerical pulse neural membrane calculation model is constructed, and a learning pulse neural membrane calculation model algorithm is designed. Finally, transformer risk assessment is realized based on the learning numerical pulse neural membrane system. The present invention can deal with complex fault modes and capture these deep-level features, and the assessment accuracy is high.

[0006] The present invention is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides a transformer risk assessment method based on a learning numerical pulse neural membrane, the method comprising:

[0008] Obtain the basic data of power transformers and construct a power transformer data set. The basic data of power transformers refers to oil chromatogram data.

[0009] Based on the basic data of power transformers, the generation mechanism of dissolved gas in transformer insulating oil is analyzed, and the relationship between dissolved gas in transformer insulating oil and risk level is analyzed to obtain the transformer status;

[0010] Construct a learning numerical pulse neural membrane computing model, and add the dimension-increasing structure and weight adaptive adjustment strategy to the learning numerical pulse neural membrane computing model;

[0011] According to the power transformer dataset, transformer risk assessment is performed based on the learning numerical pulse neural membrane calculation model to obtain the assessment result, namely the transformer status.

[0012] Furthermore, the oil chromatogram data include power transformer online monitoring data, power transformer offline experimental data and data published in other literatures.

[0013] Furthermore, the transformer status includes a normal status, a minor fault status and a major fault status, wherein:

[0014] Minor fault conditions include transformer medium and low temperature overheating, transformer partial discharge and transformer low energy discharge conditions;

[0015] Serious fault conditions include transformer overheating and transformer high-energy discharge conditions.

[0016] Furthermore, the learning numerical pulse neural membrane computing model includes an input layer, a preprocessing layer, a classification layer and an output layer; the expression of the learning numerical pulse neural membrane computing model is:

[0017] Π=(σ 1,1 ,σ 2,1 ,…,σ m,1 ,σ 1,2 ,σ 2,2 ,…,σ k,2 ,σ 1,3 ,σ 2,3 ,…,σ m,3 ,σ 1,4 ,syn,in,out)

[0018] Among them, σ 1,1 ,σ 2,1 ,…,σ m,1 Represents the m neurons in the first layer of the learning numerical spike neural membrane computing model LNSNPS;

[0019] σ 1,2 ,σ 2,2 ,...,σ k,2 Represents the k neurons in the second layer of the learning numerical spike neural membrane computing model LNSNPS;

[0020] σ 1,3 ,σ 2,3 ,…,σ n,3 Represents n neurons in the third layer of the learning numerical spike neural membrane computing model LNSNPS;

[0021] syn={(σ i ,σ j )|1≤i≤m AND 1≤j≤m AND i≠j} represents the synapse of a neuron;

[0022] in={σ 1,1 ,σ 2,1 ,…,σ m,1} is a finite set of input neurons; out = σ 1,4 is an output neuron.

[0023] Furthermore, the adaptive adjustment strategy of the dimension-raising structure and weights is added to the learning numerical pulse neural membrane computation model, including:

[0024] In the learning numerical pulse neural membrane computing model, the Maclaurin sequence method is used to obtain the high-dimensional representation of low-dimensional data;

[0025] An adaptive adjustment mechanism, namely normalized least mean square, is used to adaptively adjust the synaptic weights between the second and third layers of the learning numerical pulse neural membrane computing model.

[0026] Furthermore, the calculation formula of the adaptive adjustment mechanism is:

[0027]

[0028] in, and Represent the weights before and after updating respectively; d j is the expected output of neuron j; y j is the actual output of neuron j; x i (t) represents the state of neuron i at time t; is the adaptive step size, and c is the regularization factor used to prevent division by zero.

[0029] Furthermore, the learning process of the learning numerical pulse neural membrane computing model includes:

[0030] Step A: Divide the power transformer dataset into a training set and a test set (three quarters for training and one quarter for testing), and input the features of the sample data from the environment into the neurons of the first layer (in1 = x1, in2 = x2, ..., in m =x m );

[0031] Step B: Determine synaptic weights

[0032] Step C: Output data x from the neurons in the first layer 1,2 ,x 2,2 ,...,xk,2 =X1W1, the i-th neuron σ in the second layer is calculated i,2 The numerical variable x i,2 ,1≤i≤k, where X1 is the input data of the first layer X1=in1,in2,…,in m , W1 represents the dimension-raising matrix;

[0033] Step D: Initialize parameters: K, α, EPOCH, W, where K is the dimension-raising order, α is the learning rate, EPOCH is the number of iterations, and W is the initial weight matrix;

[0034] Step E: Perform data training:

[0035] Step F: According to the value of K, the dimension is upgraded to K order accordingly; the weight of the upgraded dimension is generated according to the pulse value entering the input layer, and the neurons in the preprocessing layer The impulse value of is calculated by the dimension-raising weight;

[0036] Step G: When t<=maxt, where t is the time step and maxt is the maximum time step, execute:

[0037] G1: According to the initialization learning weight Calculate the third layer of neurons The input value X3 = x 1,3 ,x 2,3 ,...,x n,3 ;

[0038] G2: The third layer of neurons Each neuron in contains two production functions, with a threshold T, which is determined by the neuron The pulse value at this time determines; the neurons whose pulse value can reach the threshold T will pass 1 to the output layer neurons through the output class weight The remaining neurons will pass 0 to the output layer neurons

[0039] G3: Data categories by neurons The output value of is given.

[0040] G4: Calculate the difference e between the output category and the label category, and use it to update the classification weight W1,

[0041]

[0042] Until t>maxt, the loop ends;

[0043] Step H: Save the classification model And validate on the test set.

[0044] In a second aspect, the present invention further provides a transformer risk assessment system based on a learning numerical pulse neural membrane, the system comprising:

[0045] An acquisition unit, used for acquiring basic data of power transformers and constructing a data set of power transformers, wherein the basic data of power transformers refers to oil chromatogram data;

[0046] An analysis unit is used to analyze the generation mechanism of gas dissolved in transformer insulating oil based on basic data of power transformer, and to analyze the relationship between gas dissolved in transformer insulating oil and risk level, so as to obtain transformer status;

[0047] A model building unit, used for building a learning numerical pulse neural membrane computing model, and adding a dimension-raising structure and a weight adaptive adjustment strategy to the learning numerical pulse neural membrane computing model;

[0048] The risk assessment unit is used to perform transformer risk assessment based on a learning numerical pulse neural membrane calculation model according to a power transformer data set to obtain an assessment result, i.e., the transformer status.

[0049] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned transformer risk assessment method based on the learning numerical pulse neural membrane when executing the computer program.

[0050] In a fourth aspect, the present invention further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned transformer risk assessment method based on the learning numerical pulse neural membrane is implemented.

[0051] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0052] The transformer risk assessment method and system based on the learning numerical pulse neural membrane of the present invention first analyzes the gas generation mechanism and operation characteristics in the oil, then analyzes the relationship between the dissolved gas in the transformer oil and the risk level, then constructs a learning numerical pulse neural membrane calculation model, designs a learning pulse neural membrane calculation model algorithm, and finally implements transformer risk assessment based on the learning numerical pulse neural membrane system. The present invention is based on the designed learning pulse neural membrane calculation model, which can cope with complex fault modes and capture these deep-level characteristics, and has a high assessment accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0054] Figure 1This is a flow chart of the transformer risk assessment method based on the learning numerical pulse neural membrane of the present invention;

[0055] Figure 2 It is a structural diagram of the learning numerical pulse neural membrane calculation model of the present invention;

[0056] Figure 3 It is a schematic diagram of the dimension-raising structure of the present invention;

[0057] Figure 4 This is a schematic diagram of the change in accuracy of the training set of the present invention;

[0058] Figure 5 This is a risk assessment process diagram of a transformer of a learning numerical pulse neural membrane system according to the present invention;

[0059] Figure 6 This is a structural block diagram of the transformer risk assessment system based on the learning numerical pulse neural membrane of the present invention. DETAILED DESCRIPTION

[0060] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0061] The present invention utilizes the chemical properties of transformer insulating oil to produce dissolved gas corresponding to the environment in the long-term operation environment and fault environment of the transformer itself, and designs a transformer risk assessment method based on the learning numerical pulse neural membrane calculation model based on the oil chromatogram information obtained by the transformer oil chromatogram online monitoring device and the learning numerical pulse neural membrane calculation model. The technical solution of the present invention is as follows: Figure 1 shown.

[0062] Example 1

[0063] like Figure 1 As shown, the transformer risk assessment method based on the learning numerical pulse neural membrane of the present invention comprises:

[0064] Step 1, obtaining basic data of power transformers and constructing a power transformer data set, where the basic data of power transformers refers to oil chromatogram data;

[0065] Considering that power transformers are important equipment in power infrastructure, their health status is one of the key factors affecting whether the power grid can continuously and stably supply electric energy, so they have received the focus of attention of power transmission and distribution industry practitioners and many experts and scholars at home and abroad. The transformer mainly improves the insulation strength of the transformer by immersing the winding in transformer oil. Due to its own chemical properties, the transformer insulating oil will decompose and produce dissolved gases corresponding to the gas under the action of electrical stress, thermal stress or electrothermal stress. Therefore, the oil chromatographic information obtained by the transformer oil chromatographic online monitoring device can further realize the health status evaluation and risk assessment of the transformer.

[0066] In this embodiment, the oil chromatogram data obtained by the present invention mainly comes from the online monitoring data of power transformers, the offline experimental data of power transformers, and the data published in other documents, totaling 313 pieces, and the data is divided into two parts: a training set and a test set, wherein the training set accounts for about 80%, and the remaining data is a test set. Methane / hydrogen, ethylene / ethane, acetylene / ethylene, acetylene / total hydrocarbons, ethylene / total hydrocarbons, methane / total hydrocarbons, (ethylene+methane) / total hydrocarbons are selected as the input feature quantities of the model, wherein the calculation method of total hydrocarbons is the sum of the contents of methane, ethane, ethylene, and acetylene.

[0067] Step 2: Based on the basic data of the power transformer, analyze the generation mechanism of the gas dissolved in the transformer insulating oil, and analyze the relationship between the gas dissolved in the transformer insulating oil and the risk level to obtain the transformer status;

[0068] Step 2 specifically includes:

[0069] Step 21: Analyze the generation mechanism of dissolved gas in transformer insulating oil

[0070] Usually, the windings of the transformer are immersed in transformer oil to improve the insulation strength. However, the transformer oil will decompose and produce gas under the action of electrical stress, thermal stress or combined electrical and thermal stress.

[0071] Electric stress is caused by the deterioration of the insulation of the transformer oil at a certain point under the action of the electric field energy, resulting in a strong electric field locally, which in turn causes strong electric stress at the insulation point. The energy generated by the electric stress directly acts on the insulation deterioration point, aggravating the degree of local insulation damage and making the insulation performance of the transformer oil worse, thereby enhancing the local electric field. Continuous discharge will inevitably lead to further deterioration of the transformer oil, and eventually cause the CH bond or CC bond of the transformer oil to be broken, thereby generating a large number of free radicals, which will cause the transformer oil to decompose and produce a large number of small molecular hydrocarbon gases or hydrogen.

[0072] Transformers will generate heat during long-term operation or when they fail, which will cause the temperature of the insulating oil to rise and generate thermal stress. Under the action of thermal stress, the CH bond or CC bond of the insulating oil breaks and generates hydrogen peroxide under the action of oxygen. The strong oxidizing property of hydrogen peroxide can cause the transformer oil to generate organic acids, carbon oxides and small molecular hydrocarbons while undergoing oxidation reaction. Organic acids and oxides deteriorate the performance of transformer oil, further exacerbating its deterioration process and eventually producing a large amount of hydrogen, low molecular hydrocarbon gas, X wax or other oxidation products.

[0073] Step 22: Analyze the relationship between the dissolved gas in transformer insulating oil and the risk level

[0074] When the transformer is affected by natural risks, human damage or body aging, there may be risks such as local failure, poor contact and body defects. At this time, the transformer faces thermal failure risks and electrical failure risks: thermal failure risks are caused by the transformer releasing a large amount of heat energy to the outside world and being unable to effectively dissipate heat through cooling devices or oil tanks. At this time, part of the energy will accelerate the deterioration process of the transformer oil and release hydrogen hydrocarbon gases to the outside world; electrical failure risks are caused by the degradation of the insulation strength of the insulating oil in a strong electric field environment, and can be divided into local discharge, low-energy discharge and high-energy discharge according to the discharge energy level.

[0075] According to the IEC60599-2015 standard, the categories of transformer thermal faults are usually defined as low-temperature overheating below 300°C, medium-temperature overheating between 300°C and 700°C, and high-temperature overheating above 700°C. When the transformer is overheated at low temperature, the decomposition of the transformer oil is not severe, so the main gas produced at this time is hydrogen. As the temperature rises, the cracking of the transformer oil gradually accelerates, and the proportion of alkane gases gradually increases. When the temperature is higher than 700°C, the decomposition process of the insulating oil is more intense, and C2H4 is gradually produced on the basis of CH4. The sum of the content of CH4 and C2H4 usually accounts for more than 80% of the total hydrocarbon gas. When the temperature is higher than 1000°C, a small amount of C2H2 will be produced in case of severe overheating, but it generally does not exceed 6% of the total content.

[0076] Since the energy density of transformer partial discharge is not high, the hydrocarbon gas generated by the discharge is relatively small, and the main characteristic gases are H2 and CH4. However, if it is not handled in time, it will develop into a low-energy discharge fault, which is manifested as electric sparks between different potentials or circulation between the lead and the shielding tube. The emergence of low-energy discharge means that the insulation performance of the transformer oil has further deteriorated, and the gases produced at this time are mainly C2H2 and H2. According to the Halstead gas partial pressure relationship, the generation temperature of C2H2 is usually higher than 1000℃, and the decomposition rate of the transformer oil is greatly improved compared with the partial discharge. Arc fault is a particularly serious fault, which has a significant impact on the power supply reliability of the region. When the transformer has an arc discharge induced by factors such as inter-turn, inter-layer, and inter-phase flashover of the coil or arcing of the selector switch, due to the high energy density of the arc, the transformer oil will quickly decompose and produce a large amount of gas. The gases produced at this time are mainly C2H2 and H2, followed by C2H4 and CH4.

[0077] On the basis of clarifying the mechanism of dissolved gas production when the insulating oil is threatened by the risk of transformer thermal failure and the risk of transformer electrical failure, combined with the impact of different risks on the power quality, grid reliability, economic production and people's normal production and living order in the process of transformer transmission and distribution of electric energy, the transformer status is divided into three states: normal, minor fault and serious fault. Among them, the minor fault status includes medium and low temperature overheating of the transformer, partial discharge of the transformer and low-energy discharge of the transformer. The serious fault status includes high temperature overheating of the transformer and high-energy discharge of the transformer.

[0078] In this embodiment, the transformer status obtained in step 2 includes a normal state, a minor fault state, and a major fault state, wherein:

[0079] Minor fault conditions include transformer medium and low temperature overheating, transformer partial discharge and transformer low energy discharge conditions;

[0080] Serious fault conditions include transformer overheating and transformer high-energy discharge conditions.

[0081] Step 3, constructing a learning numerical pulse neural membrane computing model, and adding a dimension-raising structure and a weight adaptive adjustment strategy to the learning numerical pulse neural membrane computing model;

[0082] In this embodiment, the learning numerical spiking neural membrane computing model (Learning numerical spiking neural P systems, LNSNPS) is as follows Figure 2 As shown in the figure, it includes input layer, preprocessing layer, classification layer and output layer; in order to improve the convergence and accuracy of the algorithm, a dimensionality-enhancing structure and an adaptive adjustment strategy for weights are also added.

[0083] Specifically, the expression of the learning numerical pulse neural membrane calculation model is:

[0084] Π=(σ 1,1 ,σ 2,1 ,…,σ m,1 ,σ 1,2 ,σ 2,2 ,…,σ k,2 ,σ 1,3 ,σ 2,3 ,…,σ m,3 ,σ 1,4 ,syn,in,out)

[0085] in,

[0086] (1)σ 1,1 ,σ 2,1 ,…,σ m,1 represents the m neurons in the first layer of the learning numerical pulse neural membrane computing model LNSNPS; where σ 1,1 ,σ 2,1 ,…,σ m,1 The form is as follows:

[0087] σ i,1 =(x i,1 ,Pr i,1 ,x i,1 (0)),1≤i≤m

[0088] In the formula,

[0089] (a)x i,1 ,1≤i≤m, is the i-th neuron σ in the first layer of LNSNPS i,1 A numerical variable. m,m∈R,is the number of neurons in the first layer of LNSNPS.

[0090] (b)Pr i,1 ,1≤i≤m, represents the i-th neuron σ in the first layer of LNSNPS i,1 The generating function Pr i,1 The specific format is as follows:

[0091] Pr i,1 :f(x i,1 )=x i,1 ,1≤i≤m

[0092] (c)x i,1 (0), 1≤i≤m, is the i-th neuron σ in the first layer of LNSNPS i,1 Initial values ​​for the numeric variables in .

[0093] (2)σ 1,2 ,σ 2,2 ,...,σk,2 represents the k neurons in the second layer of the learning numerical pulse neural membrane computing model LNSNPS; where σ 1,2 ,σ 2,2 ,...,σ k,2 The format is as follows:

[0094] σ i,1 =(x i,2 ,Pr i,2 ,x i,2 (0)),1≤i≤k

[0095] In the formula,

[0096] (a)x i,2 ,1≤i≤k, is the i-th neuron σ in the second layer of LNSNPS i,2 is a numerical variable. k, k∈R, is the number of neurons in the first layer of LNSNPS.

[0097] (b)Pr i,2 ,1≤i≤k, represents the i-th neuron σ in the second layer of LNSNPS i,2 Generating function Pr i,2 The specific format is as follows:

[0098] Pr i,2 :f(x i,2 )=x i,2 ,1≤i≤k

[0099] (c)x i,2 (0), 1≤i≤k, is the i-th neuron σ in the first layer of LNSNPS i,2 Initial values ​​for the numeric variables in .

[0100] (3)σ 1,3 ,σ 2,3 ,…,σ n,3 represents the n neurons in the third layer of the learning numerical pulse neural membrane computing model LNSNPS; where σ 1,3 ,σ 2,3 ,…,σ n,3 The form is as follows:

[0101] σ i,3 =(x i,3 ,Pr i,3 ,x i,3 (0)),1≤i≤n

[0102] In the formula,

[0103] (a)x i,3 ,1≤i≤n, is the i-th neuron σ in the third layer of LNSNPS i,3A numerical variable n, n∈R, is the number of neurons in the third layer of LNSNPS.

[0104] (b) represents the i-th neuron σ in the third layer of LNSNPS i,3 There is a finite set of production functions. and The specific format is described as follows:

[0105]

[0106] Where T = max(x 1,3 ,x 2,3 ,...,x n,3 ).

[0107] (c)x i,3 (0), 1≤i≤k, is the i-th neuron σ in the third layer of LNSNPS i,3 Initial values ​​for the numeric variables in .

[0108] (4)syn={(σ i ,σ j )|1≤i≤m AND 1≤j≤m AND i≠j} represents the synapse of the neuron; the synapse in the system is represented by L ij =((i,j),w ij ) represents the synapse between neurons and represents the weight on the synapse (i, j). It is worth noting that the weights between layers are set to different initial values. For example, the weight between the first layer and the second layer Determined by ascending order. At the same time, the weight between the second and third layers Updated by an adaptive learning mechanism. Finally, the weights between the third and fourth layers are set to

[0109] (5)in={σ 1,1 ,σ 2,1 ,…,σ m,1} is a finite set of input neurons;

[0110] (6)out=σ 1,4 is an output neuron.

[0111] It is worth noting that the data target of the learning numerical pulse neural membrane computing model LNSNPS is a numerical variable. Compared with the number of pulses, the LNSNPS of numerical variables can directly solve the different classification problems of real numbers. In addition, in the third layer, the generating functions can be divided into two categories: and T is the trigger condition. If x i,3 If it is greater than T, the function that generates the Otherwise, execute the generated function

[0112] In the learning numerical pulse neural membrane computing model LNSNPS, x ij represents the numerical variable in the i-th neuron of the j-th layer. If the numerical variable x ij Participated in generating the function f(x ij ), then x ij After calculation, it will be reset to zero, otherwise x ij will remain unchanged. If the numeric variable x ij Receives the generating function value a0 from the presynaptic neuron, and the synaptic weight of the presynaptic neuron is w0, then x ij The value will become a0w0(x ij =a0w0).

[0113] When the learning numerical spike neural membrane computing model LNSNPS is executed from one configuration to another, the configuration change is expressed by C t to C t+1 To describe. Configuration C of LNSNPS t Represents the value of all numerical variables at time t, using

[0114] C t =(x 1,1 (t),x 2,1 (t),…,x m,1 (t),x 1,2 (t),x 1,2 (t),…,x k,2 (t),x 13 (t),x 23 (t),…,x n,3 (t),x 1,4 (t)). If the running time of LNSNPS is finite, it can be expressed as Represents, where h represents the maximum number of learning steps.

[0115] In this embodiment, a strategy for adaptively adjusting the dimension-raising structure and weights is added to the learning numerical pulse neural membrane computation model, including:

[0116] In the learning numerical pulse neural membrane computing model, the Maclaurin sequence method is used to obtain the high-dimensional representation of low-dimensional data;

[0117] An adaptive adjustment mechanism, namely normalized least mean square, is used to adaptively adjust the synaptic weights between the second and third layers of the learning numerical pulse neural membrane computing model.

[0118] Specifically, the dimension-raising structure is as follows:

[0119] Since the single-layer perception algorithm cannot accurately identify the XOR problem, this is also one of the main challenges faced by the single-layer perception network. To address this problem, research has mainly focused on two aspects: the processing of input data and the increase in the number of network layers. However, as the number of network layers increases, the training time will also increase dramatically. Research on traditional pattern recognition algorithms has shown that high-dimensional information can transform nonlinear problems in low-dimensional space into linear problems. At the same time, the pattern recognition problem can be approximately considered as fitting a smooth hyperplane that can separate different types of data. Since any continuous and differentiable hyperplane can be represented by a Maclaurin sequence, this article will use the Maclaurin sequence method to obtain a high-dimensional representation of low-dimensional data such as Figure 3 .

[0120] The Maclaurin sequence method is improved by dimension increase to Dimension. The form of the Maclaurin sequence is shown below.

[0121]

[0122] Among them, k is the order of dimension increase; f(0,0,…,0) represents the deviation.

[0123] Specifically, the adaptive learning mechanism is as follows:

[0124] The adaptive adjustment mechanism, namely the normalized least mean square, is used to adaptively adjust the synaptic weights between the second and third layers of the learning numerical pulse neural membrane computing model. The calculation formula of NLMS is shown in the following formula.

[0125]

[0126] in, and Represent the weights before and after updating respectively. j is the expected output of neuron j; y j is the actual output of neuron j; x i (t) represents the state of neuron i at time t; is the adaptive step size, and c is the regularization factor used to prevent division by zero.

[0127] In this embodiment, the learning process of LNSNPS includes three sub-processes: data input, data dimension increase, data training, data testing and model saving. The input data is typical classification data. The increase of data dimension is realized between the first layer and the second layer. The synaptic weights between the second layer and the third layer are updated by NLMS. Finally, the trained model is saved and the classification accuracy Acc is output. The learning process of the learning numerical pulse neural membrane calculation model includes:

[0128] Step A: Divide the power transformer dataset into a training set and a test set (three quarters for training and one quarter for testing), and input the features of the sample data from the environment into the neurons of the first layer (in1 = x1, in2 = x2, ..., in m =x m );

[0129] Step B: Determine synaptic weights

[0130] Step C: Output data x from the neurons in the first layer 1,2 ,x 2,2 ,...,x k,2 =X1W1, the i-th neuron σ in the second layer is calculated i,2 The numerical variable x i,2 ,1≤i≤k, where X1 is the input data of the first layer X1=in1,in2,…,in m , W1 represents the dimension-raising matrix;

[0131] Step D: Initialize parameters: K, α, EPOCH, W, where K is the dimension-raising order, α is the learning rate, EPOCH is the number of iterations, and W is the initial weight matrix;

[0132] Step E: Perform data training:

[0133] Step F: According to the value of K, the dimension is upgraded to K order accordingly; the weight of the upgraded dimension is generated according to the pulse value entering the input layer, and the neurons in the preprocessing layer The impulse value of is calculated by the dimension-raising weight;

[0134] Step G: When t<=maxt, where t is the time step and maxt is the maximum time step, execute:

[0135] G1: According to the initialization learning weight Calculate the third layer of neurons The input value X3 = x 1,3 ,x 2,3 ,...,x n,3 ;

[0136] G2: The third layer of neurons Each neuron in contains two production functions, with a threshold T, which is determined by the neuron The pulse value at this time determines; the neurons whose pulse value can reach the threshold T will pass 1 to the output layer neurons through the output class weight The remaining neurons will pass 0 to the output layer neurons

[0137] G3: Data categories by neurons The output value of is given.

[0138] G4: Calculate the difference e between the output category and the label category, and use it to update the classification weight W1,

[0139]

[0140] Until t>maxt, the loop ends;

[0141] Step H: Save the classification model And validate on the test set.

[0142] Specifically, the algorithm flow of LNSNPS is shown in Table 1.

[0143]

[0144]

[0145] Step 4: According to the power transformer data set, transformer risk assessment is performed based on the learning numerical pulse neural membrane calculation model to obtain the assessment result, i.e., the transformer status.

[0146] In this embodiment, K=6 transformer states (N, T1T2, T3, PD, D1, D2) defined in the data set are used to identify the transformer state using a learning-based numerical pulse neural membrane system. The identified states are normal, mild (medium and low temperature (T1T2) and partial discharge (PD)) and severe (high temperature (T3), low energy discharge (D1) and high energy discharge (D2)). The identification process is as follows: Figure 5 shown.

[0147] Figure 4 For the learning process of the training set samples, the present invention uses methane / hydrogen, ethylene / ethane, acetylene / ethylene, acetylene / total hydrocarbons, ethylene / total hydrocarbons, methane / total hydrocarbons, (ethylene+methane) / total hydrocarbons as the input of the learning numerical pulse neural membrane system, and the three risk levels of the transformer are used as outputs. The dimensionality increase order of the learning numerical pulse neural membrane system is set to 4, the learning rate is set to 0.1, the maximum number of iterations is 1000, 80% of the overall data is used as the training set of the model, and the remaining data is used as the test set of the model. After training, it was found that the accuracy of the model on the training set was 91.5%, and on the test set was 86.7%.

[0148] In order to verify the feasibility of the method of the present invention, two cases are used to illustrate the process of transformer risk assessment based on the learning numerical pulse neural membrane calculation model. The chromatographic contents of transformer oil in the two cases are shown in Table 2.

[0149] Table 2 Case oil chromatographic content table

[0150]

[0151] Based on the evaluation and classification of the oil chromatogram data in the two cases by the learning numerical pulse neural membrane system, it can be seen that in case 1, 99.95% of the transformers subsequently had serious fault levels, and subsequent maintenance found that it was a high-temperature fault caused by poor contact of one-phase tap. In case 2, 80.02% of the transformers subsequently had serious fault levels, and during the subsequent maintenance process, the operation and maintenance personnel replaced the transformer's on-load switch, and the monitoring value of the transformer oil chromatogram returned to normal.

[0152] The present invention can cope with complex failure modes and capture these deep-level features with high evaluation accuracy.

[0153] Example 2

[0154] like Figure 6 As shown, the difference between this embodiment and embodiment 1 is that this embodiment provides a transformer risk assessment system based on a learning numerical pulse neural membrane, and the system corresponds to the transformer risk assessment method based on a learning numerical pulse neural membrane in embodiment 1 in a one-to-one manner; the system includes:

[0155] An acquisition unit, used for acquiring basic data of power transformers and constructing a data set of power transformers, wherein the basic data of power transformers refers to oil chromatogram data;

[0156] An analysis unit is used to analyze the generation mechanism of gas dissolved in transformer insulating oil based on basic data of power transformer, and to analyze the relationship between gas dissolved in transformer insulating oil and risk level, so as to obtain transformer status;

[0157] A model building unit, used for building a learning numerical pulse neural membrane computing model, and adding a dimension-raising structure and a weight adaptive adjustment strategy to the learning numerical pulse neural membrane computing model;

[0158] The risk assessment unit is used to perform transformer risk assessment based on a learning numerical pulse neural membrane calculation model according to a power transformer data set to obtain an assessment result, i.e., the transformer status.

[0159] Among them, the execution process of each unit can be executed according to the process steps of the transformer risk assessment method based on the learning numerical pulse neural membrane in Example 1, and will not be repeated one by one in this embodiment.

[0160] At the same time, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned transformer risk assessment method based on the learning numerical pulse neural membrane is implemented.

[0161] At the same time, the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned transformer risk assessment method based on the learning numerical pulse neural membrane is implemented.

[0162] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0163] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0164] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0165] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0166] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A transformer risk assessment method based on a learning numerical pulse neural membrane, characterized in that: The method includes: Acquire basic data of power transformers and construct a data set of power transformers, wherein the basic data of power transformers refers to oil chromatogram data; Based on the basic data of the power transformer, the generation mechanism of the gas dissolved in the transformer insulating oil is analyzed, and the relationship between the gas dissolved in the transformer insulating oil and the risk level is analyzed to obtain the transformer status; Constructing a learning numerical pulse neural membrane computing model, and adding a dimension-raising structure and a weight adaptive adjustment strategy to the learning numerical pulse neural membrane computing model; According to the power transformer data set, transformer risk assessment is performed based on a learning numerical pulse neural membrane calculation model to obtain an assessment result, namely, the transformer status.

2. The transformer risk assessment method based on learning numerical pulse neural membrane according to claim 1 is characterized in that: The oil chromatogram data include online monitoring data of power transformers, offline experimental data of power transformers and data published in literature.

3. The transformer risk assessment method based on learning numerical pulse neural membrane according to claim 1 is characterized in that: The transformer status includes a normal state, a minor fault state and a major fault state, wherein: The minor fault conditions include transformer medium and low temperature overheating, transformer partial discharge and transformer low energy discharge conditions; The serious fault conditions include transformer overheating and transformer high-energy discharge conditions.

4. The transformer risk assessment method based on learning numerical pulse neural membrane according to claim 1 is characterized in that: The learning numerical pulse neural membrane computing model includes an input layer, a preprocessing layer, a classification layer and an output layer; the expression of the learning numerical pulse neural membrane computing model is: P=(s 1,1 ,s 2,1 ,…,s m,1 ,s 1,2 ,s 2,2 ,…,s k,2 ,s 1,3 ,s 2,3 ,…,s m,3 ,s 1,4 ,syn,in,out) Among them, σ 1,1 ,σ 2,1 ,…,σ m,1 Represents the m neurons in the first layer of the learning numerical pulse neural membrane computing model; σ 1,2 ,σ 2,2 ,...,σ k,2 Represents the k neurons in the second layer of the learning numerical pulse neural membrane computing model; σ 1,3 ,σ 2,3 ,…,σ n,3 Represents n neurons in the third layer of the learning numerical pulse neural membrane computing model; syn={(σ i ,σ j )|1≤i≤m AND 1≤j≤m AND i≠j} represents the synapse of a neuron; in={σ 1,1 ,σ 2,1 ,…,σ m,1 } is a finite set of input neurons; out = σ 1,4 is an output neuron.

5. The transformer risk assessment method based on learning numerical pulse neural membrane according to claim 4 is characterized in that: The adaptive adjustment strategy of adding dimension-raising structure and weight to the learning numerical pulse neural membrane computing model includes: In the learning numerical pulse neural membrane computing model, the Maclaurin sequence method is used to obtain a high-dimensional representation of low-dimensional data; An adaptive adjustment mechanism, namely normalized least mean square, is used to adaptively adjust the synaptic weights between the second layer and the third layer of the learning numerical pulse neural membrane computing model.

6. The transformer risk assessment method based on learning numerical pulse neural membrane according to claim 5 is characterized in that: The calculation formula of the adaptive adjustment mechanism is: in, and Represent the weights before and after updating respectively; d j is the expected output of neuron j; y j is the actual output of neuron j; x i (t) represents the state of neuron i at time t; is the adaptive step size, and c is the regularization factor.

7. The transformer risk assessment method based on learning numerical pulse neural membrane according to claim 1 is characterized in that: The learning process of the learning numerical pulse neural membrane computing model includes: Step A: Divide the power transformer dataset into a training set and a test set, and input the features of the sample data from the environment into the neurons in1 = x1, in2 = x2, …, in1 in the first layer. m =x m ; Step B: Determine synaptic weights Step C: Output data x from the neurons in the first layer 1,2 ,x 2,2 ,...,x k,2 =X1W1, the i-th neuron σ in the second layer is calculated i,2 The numerical variable x i,2 ,1≤i≤k, where X1 is the input data of the first layer X1=in1,in2,…,in m , W1 represents the dimension-raising matrix; Step D: Initialize parameters: K, α, EPOCH, W, where K is the dimension-raising order, α is the learning rate, EPOCH is the number of iterations, and W is the initial weight matrix; Step E: Perform data training: Step F: According to the value of K, the dimension is upgraded to K order accordingly; the weight of the upgraded dimension is generated according to the pulse value entering the input layer, and the neurons in the preprocessing layer The impulse value of is calculated by the dimension-raising weight; Step G: When t<=maxt, where t is the time step and maxt is the maximum time step, execute: G1: According to the initialization learning weight Calculate the third layer of neurons The input value X3 = x 1,3 ,x 2,3 ,...,x n,3 ; G2: The third layer of neurons Each neuron in contains two production functions, with a threshold T, which is determined by the neuron The pulse value at this time determines; the neurons whose pulse value can reach the threshold T will pass 1 to the output layer neurons through the output class weight The remaining neurons will pass 0 to the output layer neurons G3: Data categories by neurons The output value of is obtained; G4: Calculate the difference e between the output category and the label category, and use it to update the classification weight W1, Until t>maxt, the loop ends; Step H: Save the classification model And validate on the test set.

8. A transformer risk assessment system based on a learning numerical pulse neural membrane, characterized in that: The system includes: An acquisition unit, used for acquiring basic data of a power transformer and constructing a data set of a power transformer, wherein the basic data of the power transformer refers to oil chromatogram data; An analysis unit, used to analyze the generation mechanism of gas dissolved in transformer insulating oil and the relationship between gas dissolved in transformer insulating oil and risk level based on the basic data of the power transformer, and obtain the transformer status; A model building unit, used to build a learning numerical pulse neural membrane computing model, and add a dimension-raising structure and a weight adaptive adjustment strategy to the learning numerical pulse neural membrane computing model; The risk assessment unit is used to perform transformer risk assessment based on the power transformer data set and the learning numerical pulse neural membrane calculation model to obtain an assessment result, that is, the transformer status.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the transformer risk assessment method based on the learning numerical pulse neural membrane as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the transformer risk assessment method based on a learning numerical pulse neural membrane as described in any one of claims 1 to 7 is implemented.

Images