Prediction method for ground air gap discharge voltage of grading ring

Abstract

The invention provides a method for predicting the discharge voltage of a grading ring to a ground air gap, and the method comprises the steps: defining an effective electric field region of the grading ring to the ground air gap, calculating the electric field distribution of the effective electric field region, extracting the electric field distribution characteristics of the effective electric field region, and constructing a voltage prediction model based on BPNN. The method comprises the steps of selecting electric field distribution characteristics on the basis of an MIC (Maximum Information Coefficient), constructing a data set, training a voltage prediction model through an ISSA algorithm (multi-strategy improved salp group algorithm) on the basis of the data set, and predicting the ground air gap discharge voltage of the grading ring on the basis of the trained voltage prediction model. According to the method, reference can be provided for safety and economical efficiency of design and installation of the converter station valve hall fitting grading ring in actual engineering, and the design reliability is improved.

Description

Technical Field

[0001] The invention relates to the technical field of calculating the discharge voltage between a grading ring and the ground, and in particular to a method for predicting the discharge voltage between a grading ring and the ground air gap. Background Art

[0002] The converter valve is a key device for DC transmission and conversion. In the converter valve hall, the grading ring is widely used to reduce electromagnetic interference and improve system stability. The grading ring forms a shielding effect by reasonably arranging the gap between it and the surrounding environment, reducing the interference of electric and magnetic fields, and ensuring the safe and reliable operation of the converter valve. In order to ensure the effectiveness of the grading ring, it is crucial to ensure the air gap discharge voltage between it and the ground. At present, there are some problems and limitations in the calculation method of the grading ring to ground discharge voltage. Existing methods are usually based on empirical formulas, simplified models or theoretical assumptions, but often fail to fully consider the complexity and diversity of the real engineering environment, resulting in insufficient accuracy and reliability of the calculation results. Therefore, it is very necessary to design a prediction method for the grading ring to ground air gap discharge voltage. Summary of the invention

[0003] In order to overcome the deficiencies of the prior art, the present invention aims to provide a method for predicting the discharge voltage of the air gap between a grading ring and the ground.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] The present invention provides a method for predicting the discharge voltage of an air gap between a grading ring and the ground, comprising:

[0006] Define the effective electric field area between the grading ring and the air gap to the ground;

[0007] Calculate the electric field distribution in the effective electric field area;

[0008] Extract the electric field distribution characteristics of the effective power plant area;

[0009] Construct voltage prediction model based on BPNN;

[0010] Based on MIC, the electric field distribution characteristics are selected, a data set is constructed, and the voltage prediction model is trained by the ISSA algorithm based on the data set;

[0011] The discharge voltage of the grading ring to the air gap is predicted based on the trained voltage prediction model.

[0012] Preferably, the effective electric field area of ​​the air gap between the grading ring and the ground is defined as follows:

[0013] The electric field distribution of the air gap between the grading ring and the ground is calculated by a two-dimensional axisymmetric model of the finite element method. The electric field strength E on the shortest path between the electrodes will present a U-shaped Ed curve, and its minimum value is E mn , the maximum value on the equalizing ring side is E mxh , the maximum value on the ground side is E mxl , the effective electric field region is defined as E ≥ E mn The area near the spherical surface, where the effective electric field area can be divided into a high-voltage area and a low-voltage area.

[0014] Preferably, the electric field distribution in the effective electric field region is calculated as follows:

[0015] The electric field distribution of the spherical gap in the effective electric field area is calculated based on the finite element method. The calculation domain is divided into multiple units. The total number of elements in the high-pressure area and the low-pressure area are set to be n and m respectively. For the i-th element in the high-pressure area and the j-th element in the low-pressure area, 1≤i≤n, 1≤j≤m, and the electric field strength is recorded as E i and E j , extract the E value and the area of ​​each element in the high-voltage and low-voltage areas of the effective electric field region.

[0016] Preferably, the electric field distribution characteristics of the effective power plant area include 29 characteristics, which are divided into three categories:

[0017] Characteristics related to electric field strength, including: the maximum value of E in high-pressure and low-pressure areas mxh and E mxl , the minimum value of E mn , the average value of E in the high pressure area and the low pressure area ah and E al , the variance E of the high-pressure area and the low-pressure area varh and E varl , the standard deviation E of the high-pressure area and the low-pressure area stdh and E sthl , the electric field distortion rate f in the high-voltage and low-voltage regions h and f l ;

[0018] Characteristics related to electric field energy include: the total energy W of high-pressure and low-pressure areas h and W l , the density of the high pressure area and the low pressure area W dh and W dl , local area E ≥ E mn +x(E mxh -E mn ) or E ≥ E mn +x(E mxl -E mn ) h7 , W h8 , Wh9 , W 17 , W 18 and W 19 , x is the proportionality coefficient, and its values ​​are 0.7, 0.8 and 0.9;

[0019] Volume-related features include: the total volume V of the high-pressure and low-pressure areas h and V l , local area E ≥ E mn +x(E mxh -E mn ) or E ≥ E mn +x(E mxl -E mn ) volume V h7 、V h8 、V h9 、V 17 、V 18 、V 19 , x is the proportional coefficient, and its values ​​are 0.7, 0.8 and 0.9.

[0020] Preferably, the electric field distribution characteristics are selected based on MIC to construct a data set, specifically:

[0021] Normalize the electric field distribution characteristics of the effective power plant area;

[0022] Calculate the MIC value between the normalized electric field distribution characteristics and the corresponding spherical gap breakdown voltage;

[0023] The electric field distribution features with larger MIC values ​​were selected to construct the data set;

[0024] Divide the dataset into test set and training set in proportion.

[0025] Preferably, the voltage prediction model is trained by an ISSA algorithm based on the data set, specifically:

[0026] The population size, maximum number of iterations and optimization dimension of the ISSA algorithm are set. According to the distribution of weights and deviations to be optimized, the Chebyshev chaotic map and the position of the salp group are initialized based on reverse learning. The voltage prediction model is trained based on the training set, and the prediction error of the training set is used as the fitness function. The fitness value of the population is calculated, the target source is determined, the leader and follower are reset, and then the target source is updated until the loop termination condition is met and the training is completed. The leader is updated by the differential evolution mechanism, and the follower is updated by the Levy flight mechanism algorithm and the nonlinear inertia weight.

[0027] The performance of the trained voltage prediction model is verified based on the test set.

[0028] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0029] The present invention provides a method for predicting the discharge voltage of the grading ring to the air gap on the ground, the method includes defining the effective electric field area of ​​the grading ring to the air gap on the ground, calculating the electric field distribution of the effective electric field area, extracting the electric field distribution characteristics of the effective power plant area, constructing a voltage prediction model based on BPNN, selecting the electric field distribution characteristics based on MIC (maximum information coefficient method), constructing a data set, training the voltage prediction model based on the data set through the ISSA algorithm (multi-strategy improved salp group algorithm), and realizing the prediction of the discharge voltage of the grading ring to the air gap on the ground based on the trained voltage prediction model. The beneficial effects of the present invention are:

[0030] 1. The model provided by the present invention establishes a database based on discharge experiments of multiple grading ring sizes and multiple gap distances, so that the prediction of the discharge voltage of the grading ring to the air gap on the ground is more accurate;

[0031] 2. The effective electric field feature set proposed in the present invention can characterize the spherical gap structure and can be used as the input parameter of the BPNN model for breakdown voltage prediction;

[0032] 3. After the electric field feature selection of MIC in the present invention, the ISSA optimized BPNN model can accurately predict the breakdown voltage of the IEC standard spherical gap;

[0033] 4. The present invention provides a reference for the safety and economy of the design and installation of the grading ring in the valve hall of the converter station in actual engineering, improves the design quality, reduces the workload, and saves a lot of costs. It can be widely used in the selection of the gap between the grading ring and the ground in the valve hall of the converter station. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0035] Figure 1 A logic block diagram provided for an embodiment of the present invention;

[0036] Figure 2 A schematic diagram of a method flow chart provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] The purpose of the present invention is to provide a method for predicting the discharge voltage of the grading ring to the air gap of the ground, which can provide a reference for the safety and economy of the design and installation of the grading ring of the valve hall hardware of the converter station in actual engineering, and improve the reliability of its design.

[0039] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] Figure 1 A logic block diagram provided for an embodiment of the present invention, Figure 2 A schematic diagram of a method flow chart provided by an embodiment of the present invention, such as Figure 1 and Figure 2 As shown, the present invention provides a method for predicting the discharge voltage of the air gap between the grading ring and the ground, comprising:

[0041] Step 100: defining an effective electric field region between the grading ring and the air gap to the ground;

[0042] Step 200: Calculate the electric field distribution in the effective electric field area;

[0043] Step 300: extracting electric field distribution characteristics of the effective power plant area;

[0044] Step 400: constructing a voltage prediction model based on BPNN;

[0045] Step 500: Select the electric field distribution features based on MIC (maximum information coefficient method), construct a data set, and train the voltage prediction model based on the data set through ISSA algorithm (multi-strategy improved salp group algorithm);

[0046] Step 600: Predict the discharge voltage of the grading ring to the air gap on the ground based on the trained voltage prediction model.

[0047] Step 100, defining the effective electric field area of ​​the grading ring to the air gap of the ground, specifically:

[0048] The electric field distribution of the air gap between the grading ring and the ground is calculated by a two-dimensional (2D) axisymmetric model of the finite element method (FEM). The electric field strength E on the shortest path between the electrodes will present a U-shaped Ed curve with a minimum value of E mn , the maximum value on the equalizing ring side is E mxh, the maximum value on the ground side is E mxl , the effective electric field region is defined as E ≥ E mn The effective electric field region can be divided into a high voltage (HV) region and a low voltage (LV) region.

[0049] In step 200, the electric field distribution in the effective electric field region is calculated, specifically:

[0050] The electric field distribution of the spherical gap in the effective electric field area is calculated based on the finite element method. The calculation domain is divided into multiple units. The total number of elements in the high-pressure area and the low-pressure area are set to be n and m respectively. For the i-th element in the high-pressure area and the j-th element in the low-pressure area, 1≤i≤n, 1≤j≤m, and the electric field strength is recorded as E i and E j , extract the E value and the area of ​​each element in the high-voltage and low-voltage areas of the effective electric field region.

[0051] In step 300, the electric field distribution characteristics of the effective power plant area include 29 characteristics with physical meaning or only mathematical meaning, which are divided into three categories:

[0052] (I) Characteristics related to electric field strength, including: the maximum value of E in high-pressure and low-pressure areas E mxh and E mxl , the minimum value of E mn , the average value of E in the high pressure area and the low pressure area ah and E al , the variance E of the high-pressure area and the low-pressure area varh and E varl , the standard deviation E of the high-pressure area and the low-pressure area stdh and E sthl , the electric field distortion rate f in the high-voltage and low-voltage regions h and f l ;

[0053] (ii) Characteristics related to electric field energy, including: the total energy W of the high-pressure and low-pressure areas h and W l , the density of the high pressure area and the low pressure area W dh and W dl , local area E ≥ E mn +x(E mxh -E mn ) or E ≥ E mn +x(E mxl -E mn ) h7 , W h8 , W h9 , W 17 , W 18 and W 19, x is the proportionality coefficient, with values ​​of 0.7, 0.8 and 0.9, which has no physical meaning and is only used to characterize the distribution of electric field energy;

[0054] (III) Volume-related features, including: the total volume V of the high-pressure and low-pressure areas h and V l , local area E ≥ E mn +x(E mxh -E mn ) or E ≥ E mn +x(E mxl -E mn ) volume V h7 、V h8 、V h9 、V 17 、V 18 、V 19 , x is the proportionality coefficient, and its values ​​are 0.7, 0.8 and 0.9. It has no physical meaning and is only used to characterize the distribution of electric field energy.

[0055] In step 400, a voltage prediction model is constructed based on BPNN, specifically:

[0056] The input of BPNN (Back Propagation Neural Network) can be described as: Where p is the number of the input variable, x i is the i-th input variable, w ij is the weight of the connection between the input layer and the hidden layer, θ j It is a neuron h j The bias of the neuron h j The output of j It can be calculated by the transfer function f, which is:

[0057]

[0058] Taking the output of all neurons in the hidden layer as the input of the output layer, it can be written as:

[0059]

[0060] Among them, v j is the connection weight between the hidden layer and the output layer, r is the bias of the output layer, and the final output y can be calculated as:

[0061]

[0062] The final output is compared with the target output, and the difference between the two is called the network output error;

[0063] Based on the back propagation learning algorithm, the weights and biases of the BP neural network are continuously modified to minimize the error.

[0064] In step 500, based on the MIC, the electric field distribution characteristics are selected to construct a data set, specifically:

[0065] Normalize the electric field distribution characteristics of the effective power plant area;

[0066] Calculate the MIC value between the normalized electric field distribution characteristics and the corresponding sphere gap breakdown voltage;

[0067] Select the electric field distribution characteristics with larger MIC values to construct a data set;

[0068] Divide the data set into a test set and a training set according to a ratio;

[0069] Among them, in order to calculate the MIC, the variables X and Y are randomly divided into scatterers on the horizontal and vertical axes using a grid of size m×n. The mutual information is calculated according to the distribution of the scatterers, and the maximum mutual information under each division is normalized to the range [0,1]. Since there are multiple discretization methods for the grid, the maximum normalized mutual information under different values of m and n is taken as the final MIC, which can be expressed as:

[0070]

[0071] Among them, m×n < B is the constraint condition for grid division, B is usually set to 0.6 times the data volume, I*(x,y) is the maximum mutual information between x and y under a specified division method, and lgmin(m,n) is the basic value for normalization.

[0072] In step 500, based on the data set, the voltage prediction model is trained by the ISSA algorithm, specifically:

[0073] Set the population size, maximum number of iterations, and optimization dimension of the ISSA algorithm. According to the distribution of the weights and biases to be optimized, introduce the Chebyshev chaotic map and reverse learning-based method to initialize the positions of the salp group. Train the voltage prediction model based on the training set, and use the prediction error of the training set as the fitness function to calculate the fitness values of the population, determine the target source, reset the leader and follower, and then update the target source until the loop termination condition is met to complete the training. Among them, the differential evolution mechanism is used to update the leader, and the Levy flight mechanism algorithm and non-linear inertia weight are introduced to update the follower;

[0074] The update process of the leader is as follows:

[0075]

[0076] Among them, is the position of the jth dimension in the i-th leader (i=1,2,…), F j is the position of the jth element of the target source, ub j lb j are the upper and lower bounds of the j-th dimension search space, c2 and c3 are random parameters between [0,1], and c1 is an important parameter for balancing global exploration and local exploitation. l is the current iteration, L is the maximum number of iterations, and Follower moves in sequence according to the chain rule. Its position is determined by the position of the previous iteration and the adjacent salp, that is:

[0077]

[0078] It should be noted that the present invention improves the ISSA algorithm:

[0079] 1. Since SSA uses a random method to initialize the population, the search process cannot stably obtain the target accuracy and convergence speed. The target source driven search in the leader can accelerate the convergence of the algorithm, but there are problems of insufficient traversal and insufficient global exploration. When the leader falls into a local optimum, a refined search for followers around the local optimal area will cause the search to be unable to jump out of the local optimum. To address the above problems, the ISSA in the present invention introduces the Chebyshev chaotic map and initializes the salp group based on reverse learning on the basis of the SSA algorithm to obtain a high-quality population. The Chebyshev chaotic map is defined as:

[0080] p i+1 =cos[icos -1 (p i )]

[0081] Among them, p i Between [0,1], the Chebyshev chaotic sequence is mapped to the search space of the initial solution x, and its inverse solution is calculated. The inverse solution is defined as:

[0082]

[0083] 2. In order to better play the guiding role of the leader, the present invention adopts the differential evolution mechanism to update the leader through mutation, crossover, selection and other methods to discover excellent individuals. In this way, the information exchange between leaders can be strengthened, thereby avoiding the disadvantage of targeting a single target source. Each leader x i Randomly select three different leaders and combine them together to generate variant individuals v i , which can be expressed as:

[0084]

[0085] In the formula, r1, r2, r3∈[1,0.5N] are the numbers of three different differential individuals, the proportional factor s∈[0,1] is a random number, and the new individual u i is constructed through a crossover operation:

[0086]

[0087] Wherein, rand is a random number in the interval [0,1], CR∈[0,1] is a key parameter that determines the degree of crossover. In the present invention, CR is set to 0.5, and the fitness values ​​of the original salp and the new salp after the crossover operation are calculated:

[0088]

[0089] In order to avoid excessive reliance on the previous salp and prevent falling into the local optimum, the present invention introduces the Levy flight mechanism algorithm and the nonlinear inertia weight to refine the follower update method. The update formula of the new follower position is:

[0090]

[0091] Among them, Levy(k) can generate a variety of random step sizes, fit is the overall average fitness value, w1 is the initial weight, w2 is the termination weight at the end of the iteration, w1 and w2 are taken as 0.4 and 0.8 respectively, and the inertia weight decreases nonlinearly from 0.8 to 0.4. A larger inertia weight at the beginning of the iteration maintains the algorithm's exploration ability, and a smaller inertia weight at the end of the iteration helps the algorithm converge quickly.

[0092] The ISSA algorithm based on the improved SSA proposed in the present invention can further improve the diversity of the population, coordinate the salp population, and complete the overall exploration and local development in a more balanced way by introducing a chaotic elite group initialization mechanism, a leader differential evolution mechanism, and a follower update mechanism based on nonlinear inertia weight.

[0093] The performance of the trained voltage prediction model is verified based on the test set.

[0094] In step 600, the prediction of the discharge voltage of the grading ring to the air gap is realized based on the trained voltage prediction model, specifically:

[0095] The ultimate goal of model training is to optimize weights and biases, and then use the trained model to make predictions.

[0096] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0097] The present invention uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting the discharge voltage of the air gap between a grading ring and the ground, characterized in that: include: Define the effective electric field area between the grading ring and the air gap to the ground; Calculate the electric field distribution in the effective electric field area; Extract the electric field distribution characteristics of the effective power plant area; Construct voltage prediction model based on BPNN; Based on MIC, the electric field distribution characteristics are selected, a data set is constructed, and the voltage prediction model is trained by the ISSA algorithm based on the data set; The discharge voltage of the grading ring to the air gap is predicted based on the trained voltage prediction model.

2. The method according to claim 1, characterized in that The effective electric field area of ​​the grading ring to the air gap is defined as: The electric field distribution of the air gap between the grading ring and the ground is calculated by a two-dimensional axisymmetric model of the finite element method. The electric field strength E on the shortest path between the electrodes will present a U-shaped Ed curve, and its minimum value is E mn , the maximum value on the equalizing ring side is E mxh , the maximum value on the ground side is E mxl , the effective electric field region is defined as E ≥ E mn The area near the spherical surface, where the effective electric field area can be divided into a high-voltage area and a low-voltage area.

3. The method according to claim 1, characterized in that Calculate the electric field distribution in the effective electric field area, specifically: The electric field distribution of the spherical gap in the effective electric field area is calculated based on the finite element method. The calculation domain is divided into multiple units. The total number of elements in the high-pressure area and the low-pressure area are set to be n and m respectively. For the i-th element in the high-pressure area and the j-th element in the low-pressure area, 1≤i≤n, 1≤j≤m, and the electric field strength is recorded as E i and E j , extract the E value and the area of ​​each element in the high-voltage and low-voltage areas of the effective electric field region.

4. The method according to claim 1, characterized in that: The electric field distribution characteristics of the effective power plant area include 29 characteristics, which are divided into three categories: Characteristics related to electric field strength, including: the maximum value of E in high-pressure and low-pressure areas mxh and E mxl , the minimum value of E mn , the average value of E in the high pressure area and the low pressure area ah and E al , the variance E of the high-pressure area and the low-pressure area varh and E varl , the standard deviation E of the high-pressure area and the low-pressure area stdh and E sthl , the electric field distortion rate f in the high-voltage and low-voltage regions h and f l ; Characteristics related to electric field energy include: the total energy W of high-pressure and low-pressure areas h and W l , the density of the high pressure area and the low pressure area W dh and W dl , local area E ≥ E mn +x(E mxh -E mn ) or E ≥ E mn +x(E mxl -E mn ) h7 , W h8 , W h9 , W 17 , W 18 and W 19 , x is the proportionality coefficient, and its values ​​are 0.7, 0.8 and 0.9; Volume-related features include: the total volume V of the high-pressure and low-pressure areas h and V l , local area E ≥ E mn +x(E mxh -E mn ) or E ≥ E mn +x(E mxl -E mn ) volume V h7 、V h8 、V h9 、V 17 、V 18 、V 19 , x is the proportional coefficient, and its values ​​are 0.7, 0.8 and 0.

9.

5. The method according to claim 1, characterized in that Based on MIC, the electric field distribution characteristics are selected and the data set is constructed, specifically: Normalize the electric field distribution characteristics of the effective power plant area; Calculate the MIC value between the normalized electric field distribution characteristics and the corresponding spherical gap breakdown voltage; The electric field distribution features with larger MIC values ​​were selected to construct the data set; Divide the dataset into test set and training set in proportion.

6. The method according to claim 1, characterized in that The voltage prediction model is trained based on the data set using the ISSA algorithm, specifically: The population size, maximum number of iterations and optimization dimension of the ISSA algorithm are set. According to the distribution of weights and deviations to be optimized, the Chebyshev chaotic map and the position of the salp group are initialized based on reverse learning. The voltage prediction model is trained based on the training set, and the prediction error of the training set is used as the fitness function. The fitness value of the population is calculated, the target source is determined, the leader and follower are reset, and then the target source is updated until the loop termination condition is met and the training is completed. The leader is updated by the differential evolution mechanism, and the follower is updated by the Levy flight mechanism algorithm and the nonlinear inertia weight. The performance of the trained voltage prediction model is verified based on the test set.

Images