CN111200290B - Intelligent control method of phase change switch for three-phase unbalance treatment of transformer area - Google Patents

Abstract

The invention provides an intelligent control method of a commutation switch for the treatment of three-phase unbalance of a transformer area, which is characterized by adopting a K-means algorithm to classify loads, generating a sample set, constructing a Deep Belief Network (DBN), deeply excavating a complex association relation between input characteristic parameters and the state of the commutation switch of the transformer area, training the deep belief network by using the sample set to obtain a transformer area commutation switch control model and finally applying the intelligent control strategy of the commutation switch. The invention provides an intelligent control method of a phase change switch for three-phase unbalance management of a transformer area, which takes a transient current curve as a load classification basis, preferentially selects the phase change switch corresponding to a load with smaller transient current amplitude when switching the load, and ensures the safe and stable operation of the transformer area while managing the three-phase unbalance of the transformer area; and (3) excavating the potential deep relation between the operation condition of the transformer area and the optimal commutation strategy by utilizing a deep learning technology to form a discovery mechanism of the optimal commutation strategy of the transformer area, so as to realize the quick determination of the control strategy of the commutation switch.

Description

Intelligent control method of phase change switch for three-phase unbalance treatment of transformer area

Technical Field

The invention relates to the technical field of three-phase unbalance treatment of transformer areas, in particular to an intelligent control method of a phase change switch for three-phase unbalance treatment of the transformer areas.

Background

A large number of low-voltage distribution substations exist in China, are distributed in various cities throughout the country, and are mostly supplied with three-phase power in a power supply mode. The user characteristics determine that the administered loads of the low-voltage distribution station area are mostly single-phase loads, and a small amount of three-phase loads exist at the same time. Because load access inequality, user's power consumption characteristics have reasons such as difference, each looks load in low voltage distribution station district is in unbalanced state most of the time for there is the unbalanced three phase phenomenon in whole low voltage distribution station district. The three-phase imbalance can bring many problems to the transformer area, such as the increase of the loss of a transformer and a line, the reduction of heavy-load phase voltage and the increase of light-load phase voltage, thereby causing adverse effects on electric equipment, the reduction of the operation efficiency and the overload capacity of the transformer, the increase of the eddy current loss of the transformer, the increase of the operation temperature, the reduction of the service life and the like. Therefore, the three-phase unbalance treatment of the transformer area is an important work and is a necessary condition for realizing the economic and safe operation of the transformer area.

There are various three-phase imbalance management methods in the distribution room, wherein the methods of manual phase sequence adjustment, reactive compensation, three-phase load asymmetric adjustment and compensation, etc. have been gradually stopped being used due to the reasons of insufficient flexibility, high equipment and labor cost, etc., and the most common three-phase imbalance management measure at present is to install a low-voltage load phase-change switch. The commutation switch is usually installed on the low-voltage side of the transformer, and can monitor the line current and flexibly switch the load among A, B, C three phases. When the three-phase unbalance degree of the transformer area exceeds a limit value, the load access phase is adjusted through the low-voltage load phase change switch, and the purpose of treatment is achieved. The establishment of the commutation strategy is the core content of the scheme, and the evaluation indexes of the commutation strategy generally comprise commutation effect, commutation speed and equipment loss, wherein the commutation effect and the equipment loss cannot be optimal at the same time. Various methods are used for making a commutation strategy, such as an intelligent algorithm, a linear programming method, a dynamic programming method and the like, but the problems that the solving efficiency is reduced along with the increase of the number of switches cannot be fundamentally solved.

Disclosure of Invention

The invention provides an intelligent control method of a phase change switch for the three-phase unbalance management of a transformer area, aiming at overcoming the technical defect that the solving efficiency is reduced when the number of the phase change switches is increased in the conventional three-phase unbalance management method of the transformer area.

In order to solve the technical problems, the technical scheme of the invention is as follows:

an intelligent control method of a phase change switch for three-phase unbalance treatment of a transformer area comprises the following steps:

s1: taking a phase-change transient current curve generated by loads connected with different phase-change switches during phase change as a basis, and performing clustering analysis on the phase-change transient current curves of different loads by adopting a K-means algorithm, so that N phase-change switches are divided into K types according to the phase-change transient current curves of the loads connected with the phase-change switches and are sorted;

s2: counting the current value I of each line of N phase change switches installed under the distribution room₁、I₂、……、I_NA, B, C three-phase current value I of outgoing line of sum region_A、I_B、I_CHistorical data of (a); selecting a plurality of groups of historical data, formulating a commutation switch control strategy, and forming a commutation switch control strategy sample set for subsequent deep learning model training;

s3: dividing the sample set data generated in S2 into a training sample set and a test sample set;

s4: building a deep belief network DBN, and constructing I_A、I_B、I_CAnd I₁、I₂、…、I_NAs an input to the DBN; the output of the DBN is the action mark of each commutation switch: chi shape_i，i＝1,2,L,N；

S5: acquiring initial structure parameters of the DBN, wherein the initial structure parameters comprise the number of nodes of an input display layer and an output display layer, the number of hidden layers and the number of hidden layer nodes; inputting the training sample set, performing unsupervised layer-by-layer pre-training on the DBN by adopting a greedy learning algorithm, and finally updating the interlayer connection weight W, the bias b of a visible layer and the bias c of a hidden layer of each restricted Boltzmann machine RBM to complete the pre-training process of the DBN;

s6: adopting a gradient descent BP algorithm to perform supervised integral fine adjustment on a mean square error function L (theta) of the DBN after pre-training; after the DBN is fully trained, an intelligent control strategy model of the phase change switch of the transformer area to be verified is obtained;

s7: inputting the test sample set into a model to be verified, checking whether an error value U between a phase change switch state output value and an actual value is within a preset error range, if not, returning to S5, and continuing training to improve the model precision; and if so, obtaining a platform area phase change switch control model for making an on-line decision and controlling the platform area phase change switch.

Preferably, the K-means algorithm specifically comprises the following steps:

s1.1: setting the clustering number of the commutation transient current curve as k, and initializing k clustering centers at random to obtain an initial clustering center; inputting the commutation transient current curve of the load connected with N commutation switches as sample data X_j，j＝1,2,L,N；

S1.2: calculating X_jDistance from the center of each cluster, X_jAnd (3) classifying the cluster centers with the closest distance into the corresponding classes, wherein the calculation formula is as follows:

dist(ω_i,X_j)＝||X_j-ω_i||²

in the formula, ω_iWhere i is 1,2, L, k denotes the ith cluster center, dist (ω)_i,X_j) Representing the distance between the clustering center and the sample data;

s1.3: computing X in each class_jThe calculation result is used as a new clustering center, and the calculation formula is as follows:

s1.4: and repeating S1.2 and S1.3 until the clustering center is not changed or the change is very small, so that the commutation switch is divided into k types according to the commutation transient current curve of the load connected with the commutation switch.

Preferably, in the sorting rule in S1, the commutation switch with the smallest transient current amplitude is located at the front end of the sequence, and the intra-class commutation switches are sorted in ascending order according to the magnitude of the transient current amplitude.

Preferably, the step of formulating the commutation switch control strategy by S2 specifically includes the following steps:

s2.1: initializing system parameters, and enabling t to be 0 and num to be 0, wherein t is a system timer, and num is the number of times of overrun of the three-phase current unbalance degree; setting a three-phase current monitoring period T, and setting a maximum allowable value delta and an allowable overrun number n of the three-phase current unbalance;

s2.2: when T is T, calculating average current I of the transformer area_aveAnd three-phase unbalance eta, and the calculation formula is as follows:

s2.3: judging whether eta exceeds delta or not, if not, resetting t to be 0, and returning to S2.2; if yes, executing S2.4;

s2.4: a, B, C calculating the difference delta I between the three-phase current and the average current_A、ΔI_B、ΔI_CThe calculation formula is as follows:

ΔI_A＝I_A-I_ave

ΔI_B＝I_B-I_ave

ΔI_C＝I_C-I_ave；

s2.5: controlling the commutation switch to carry out load adjustment according to the calculation result of the S2.4;

s2.6: judging whether the eta after phase conversion is lower than delta, if not, continuing to execute S2.5; otherwise, recording each phase current I of the phase conversion foreground area_A、I_B、I_CCurrent I corresponding to each phase change switch before phase change₁、I₂、……、I_NEach commutation switch action mark x_iForming a commutation switch control strategy sample, resetting t to 0, and returning to step S2.2.

Preferably, the specific adjusting method in S2.5 is as follows:

if I_A>I_B>I_CSelecting the switch with the connected load in the A phase from the class 1 commutation switch to the C phase, wherein the switching amount is

If the switchable quantity of the A-phase load in the class 1 phase change switch is insufficient, selecting the class 2 phase change switch to continue switching, and repeating the operation until the switchable quantity is not less than delta I;

if I_A>I_B＝I_CSelecting the switch with the connected load in the A phase from the class 1 commutation switch to the B phase, wherein the switching amount is

If the switchable quantity of the A-phase load in the class 1 phase change switch is insufficient, selecting the class 2 phase change switch to continue switching, and repeating the operation until the switchable quantity is not less than delta I;

if I_A＝I_B>I_CSelecting the switch with the connected load in the A phase from the class 1 commutation switch to the C phase, wherein the switching amount is

If the switchable quantity of the A-phase load in the class 1 phase change switch is insufficient, the class 2 phase change switch is selected to continue switching, and the like is repeated until the switchable quantity is not less than delta I.

Preferably, in each phase change switch operation flag of S2.6, 1 indicates switching from phase a to phase B, 2 indicates switching from phase a to phase C, 3 indicates switching from phase B to phase a, 4 indicates switching from phase B to phase C, 5 indicates switching from phase C to phase a, and 6 indicates switching from phase C to phase B.

Preferably, S2.6 each commutation switch action mark χ_iE {1,2,3,4,5,6}, 1 denotes a switch from a phase to B phase, 2 denotes a switch from a phase to C phase, 3 denotes a switch from B phase to a phase, 4 denotes a switch from B phase to C phase, 5 denotes a switch from C phase to a phase, and 6 denotes a switch from C phase to B phase.

Preferably, the ratio of the training sample set to the test sample set in S3 is 8: 2.

Preferably, in S5, the number of input layer nodes of the DBN is equal to (N +3) of the three-phase unbalanced electrical characteristic parameters m in the distribution area, that is, [ I [ ]_A,I_B,I_C,I₁,I₂,……I_N](ii) a The output layer corresponds to the action states of N phase change switches, and the number of nodes is N; the structure parameter of the initialized deep belief network is 3 hidden layers, and the number of the nodes of the initial hidden layers is s-2 m + 1.

Preferably, the greedy learning algorithm in S5 specifically includes the following steps:

s5.1, changing x to [ I_A,I_B,I_C,I₁,I₂,……I_N]Given to the input display layer, the probability P (h) of each hidden layer neuron h being activated is calculated by the following formula₁|v₁):

P(h_j|v)＝σ(b_j+∑_iW_i,jx_i)

In the formula, coefficient b_jBias for the jth apparent neuron, W_i,jThe weight value between any two connected neurons i and j is sigma function;

s5.2, extracting a sample h from the calculated probability distribution by adopting Gibbs sampling₁：

h₁～P(h₁|v₁)；

S5.3 use h₁Reconstructing the visualization layer, namely reversely deducing the visualization layer through the hidden layer, and calculating the probability P (v) that each visualization layer neuron v is activated by using the following formula₂|h₁)；

P(v_i|h)＝σ(c_i+∑_jW_i,jh_j)

In the formula, coefficient c_iA bias for the ith hidden layer neuron;

s5.4: similarly, a sample v is extracted from the calculated probability distribution using Gibbs sampling₂By v₂Calculating the activated probability of each neuron in the hidden layer again to obtain a probability score P (h)₂|v₂)：

v₂～P(v₂|h₁)；

S5.5: update the weight W with the contrast divergence method:

W←W+λ(P(h₁|v₁)v₁-P(h₂|v₂)v₂)

b←b+λ(v₁-v₂)

c←c+λ(h₁-h₂)

in the formula, λ is a learning rate of the contrast divergence algorithm.

Preferably, the supervised global refinement is a top-down callback on the pre-trained DBN, and updates the network parameter vector θ to [ b, c, W ]:

U_i＝y_i-y_out

in the formula, alpha is the learning rate of BP algorithm, U is the error value, n_outThe number of output nodes; y is_iTo train the actual output value of the sample, y_outIs the predicted output value of the DBN.

Compared with the prior art, the technical scheme of the invention has the beneficial effects that:

the invention provides an intelligent control method of a phase change switch for three-phase unbalance management of a transformer area, which takes a transient current curve as a load classification basis, preferentially selects the phase change switch corresponding to a load with smaller transient current amplitude when switching the load, and ensures the safe and stable operation of the transformer area while managing the three-phase unbalance of the transformer area; and (3) excavating the potential deep relation between the operation condition of the transformer area and the optimal commutation strategy by utilizing a deep learning technology to form a discovery mechanism of the optimal commutation strategy of the transformer area, so as to realize the quick determination of the control strategy of the commutation switch.

Drawings

FIG. 1 is a flow chart of the implementation of the technical solution of the present invention.

Detailed Description

The drawings are for illustrative purposes only and are not to be construed as limiting the patent;

for the purpose of better illustrating the embodiments, certain features of the drawings may be omitted, enlarged or reduced, and do not represent the size of an actual product;

it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.

The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.

As shown in fig. 1, an intelligent control method for a phase change switch for three-phase imbalance management in a transformer area includes the following steps:

s1: taking a phase-change transient current curve generated by loads connected with different phase-change switches during phase change as a basis, and performing clustering analysis on the phase-change transient current curves of different loads by adopting a K-means algorithm, so that N phase-change switches are divided into K classes according to the phase-change transient current curves of the loads connected with the phase-change switches and are sorted; the phase change switches with the minimum transient current amplitude are positioned at the front end of the sequence, and the phase change switches in the class are also sorted in an ascending order according to the transient current amplitude; the K-means algorithm comprises the following specific steps:

s1.1: setting the clustering number of the commutation transient current curve as k, and initializing k clustering centers at random to obtain an initial clustering center; inputting the commutation transient current curve of the load connected with N commutation switches as sample data X_j，j＝1,2,L,N；

S1.2: calculating X_jDistance from the center of each cluster, X_jAnd (3) classifying the cluster centers with the closest distance into the corresponding classes, wherein the calculation formula is as follows:

dist(ω_i,X_j)＝||X_j-ω_i||

in the formula, ω_iWhere i is 1,2, L, k denotes the ith cluster center, dist (ω)_i,X_j) Representing the distance between the clustering center and the sample data;

s1.3: computing X in each class_jTaking the calculated result as a new clustering center, countingThe calculation formula is as follows:

s1.4: and repeating S1.2 and S1.3 until the clustering center is not changed or the change is very small, so that the commutation switch is divided into k types according to the commutation transient current curve of the load connected with the commutation switch.

S2: counting the current value I of each line of N phase change switches installed under the distribution room₁、I₂、……、I_NA, B, C three-phase current value I of outgoing line of sum region_A、I_B、I_CHistorical data of (a);

s2.1: initializing system parameters, and enabling t to be 0 and num to be 0, wherein t is a system timer, and num is the number of times of overrun of the three-phase current unbalance degree; setting a three-phase current monitoring period T, and setting a maximum allowable value delta and an allowable overrun number n of the three-phase current unbalance;

s2.2: when T is T, calculating average current I of the transformer area_aveAnd three-phase unbalance eta, and the calculation formula is as follows:

s2.3: judging whether eta exceeds delta or not, if not, resetting t to be 0, and returning to S2.2; if yes, executing S2.4;

s2.4: a, B, C calculating the difference delta I between the three-phase current and the average current_A、ΔI_B、ΔI_CThe calculation formula is as follows:

ΔI_A＝I_A-I_ave

ΔI_B＝I_B-I_ave

ΔI_C＝I_C-I_ave；

s2.5: controlling the commutation switch to carry out load adjustment according to the calculation result of the S2.4; the specific adjusting method comprises the following steps:

if I_A>I_B>I_CSelecting the switch with the connected load in the A phase from the class 1 commutation switch to the C phase, wherein the switching amount is

If the switchable quantity of the A-phase load in the class 1 phase change switch is insufficient, selecting the class 2 phase change switch to continue switching, and repeating the operation until the switchable quantity is not less than delta I;

if I_A>I_B＝I_CSelecting the switch with the connected load in the A phase from the class 1 commutation switch to the B phase, wherein the switching amount is

If the switchable quantity of the A-phase load in the class 1 phase change switch is insufficient, selecting the class 2 phase change switch to continue switching, and repeating the operation until the switchable quantity is not less than delta I;

if I_A＝I_B>I_CSelecting the switch with the connected load in the A phase from the class 1 commutation switch to the C phase, wherein the switching amount is

If the switchable quantity of the A-phase load in the class 1 phase change switch is insufficient, the class 2 phase change switch is selected to continue switching, and the like is repeated until the switchable quantity is not less than delta I.

S2.6: judging whether the eta after phase conversion is lower than delta, if not, continuing to execute S2.5; otherwise, recording each phase current I of the phase conversion foreground area_A、I_B、I_CCurrent I corresponding to each phase change switch before phase change₁、I₂、……、I_NEach commutation switch action mark x_iI is 1,2, L, N, forming a commutation switch control strategy sample, resetting t is 0, and returning to step S2.2.

Wherein, each commutation switch action mark χ_iE.g. {1,2,3,4,5,6}, where 1 denotes a switch from A phase to B phase, 2 denotes a switch from A phase to C phase,3 denotes switching from phase B to phase a, 4 denotes switching from phase B to phase C, 5 denotes switching from phase C to phase a, and 6 denotes switching from phase C to phase B.

Selecting a plurality of groups of historical data, formulating a commutation switch control strategy, and forming a commutation switch control strategy sample set for subsequent deep learning model training.

S3: dividing the sample set data generated in S2 into a training sample set and a test sample set;

s4: building a deep belief network DBN, and constructing I_A、I_B、I_CAnd I₁、I₂、…、I_NAs an input to the DBN; the output of the DBN is the action mark of each commutation switch: chi shape_i；

The number of input layer nodes of the DBN is equal to (N +3) of three-phase unbalanced electrical characteristic parameters m of the transformer area, namely [ I [ ]_A,I_B,I_C,I₁,I₂,……I_N](ii) a The output layer corresponds to the action states of N phase change switches, and the number of nodes is N; the structure parameter of the initialized deep belief network is 3 hidden layers, and the number of the nodes of the initial hidden layers is s-2 m + 1.

S5: acquiring initial structure parameters of the DBN, wherein the initial structure parameters comprise the number of nodes of an input display layer and an output display layer, the number of hidden layers and the number of hidden layer nodes; inputting the training sample set and carrying out unsupervised layer-by-layer pre-training on the DBN by adopting a greedy learning algorithm; the greedy learning algorithm specifically comprises the following steps:

s5.1, changing x to [ I_A,I_B,I_C,I₁,I₂,……I_N]Given to the input display layer, the probability P (h) of each hidden layer neuron h being activated is calculated by the following formula₁|v₁):

P(h_j|v)＝σ(b_j+∑_iW_i,jx_i)

In the formula, coefficient b_jBias for the jth apparent neuron, W_i,jThe weight value between any two connected neurons i and j is sigma function;

s5.2 extracting one Gibbs sample from the calculated probability distributionSample h₁：

h₁～P(h₁|v₁)；

S5.3 use h₁Reconstructing the visualization layer, namely reversely deducing the visualization layer through the hidden layer, and calculating the probability P (v) that each visualization layer neuron v is activated by using the following formula₂|h₁)；

P(v_i|h)＝σ(c_i+∑_jW_i,jh_j)

In the formula, coefficient c_iA bias for the ith hidden layer neuron;

s5.4: similarly, a sample v is extracted from the calculated probability distribution using Gibbs sampling₂By v₂Calculating the activated probability of each neuron in the hidden layer again to obtain a probability score P (h)₂|v₂)：

v₂～P(v₂|h₁)；

S5.5: update the weight W with the contrast divergence method:

W←W+λ(P(h₁|v₁)v₁-P(h₂|v₂)v₂)

b←b+λ(v₁-v₂)

c←c+λ(h₁-h₂)

in the formula, λ is a learning rate of the contrast divergence algorithm.

Finally updating the interlayer connection weight W, the bias b of the visible layer and the bias c of the hidden layer of each limited Boltzmann machine RBM to complete the pre-training process of the DBN;

s6: performing supervised integral fine adjustment on a mean square error function L (theta) of the pre-trained DBN by adopting a gradient descent BP algorithm, performing top-down callback on the pre-trained DBN, and updating a network parameter vector theta to be [ b, c, W ]:

U_i＝y_i-y_out

in the formula, alpha is the learning rate of BP algorithm, U is the error value, n_outThe number of output nodes; y is_iTo train the actual output value of the sample, y_outIs the predicted output value of the DBN.

After the DBN is fully trained, an intelligent control strategy model of the phase change switch of the transformer area to be verified is obtained;

s7: inputting the test sample set into a model to be verified, checking whether an error value U between a phase change switch state output value and an actual value is within a preset error range, if not, returning to S5, and continuing training to improve the model precision; and if so, obtaining a platform area phase change switch control model for making an on-line decision and controlling the platform area phase change switch.

It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.

Claims (9)

1. An intelligent control method of a phase change switch for three-phase unbalance treatment of a transformer area is characterized by comprising the following steps:

s1: taking a phase-change transient current curve generated by loads connected with different phase-change switches during phase change as a basis, and performing clustering analysis on the phase-change transient current curves of different loads by adopting a K-means algorithm, so that N phase-change switches are divided into K types according to the phase-change transient current curves of the loads connected with the phase-change switches and are sorted;

s2: counting the current value I of each line of N phase change switches installed under the distribution room₁、I₂、……、I_NAnd outlet of the stage areaA. B, C three-phase current value I_A、I_B、I_CHistorical data of (a); selecting a plurality of groups of historical data, formulating a commutation switch control strategy, and forming a commutation switch control strategy sample set for subsequent deep learning model training;

s3: dividing the sample set data generated in S2 into a training sample set and a test sample set;

s4: building a deep belief network DBN, and constructing I_A、I_B、I_CAnd I₁、I₂、…、I_NAs an input to the DBN; the output of the DBN is the action mark of each commutation switch: chi shape_i，i＝1,2,…,N；

S5: acquiring initial structure parameters of the DBN, wherein the initial structure parameters comprise the number of nodes of an input display layer and an output display layer, the number of hidden layers and the number of hidden layer nodes; inputting the training sample set, performing unsupervised layer-by-layer pre-training on the DBN by adopting a greedy learning algorithm, and finally updating the interlayer connection weight W, the bias b of a visible layer and the bias c of a hidden layer of each restricted Boltzmann machine RBM to complete the pre-training process of the DBN;

the greedy learning algorithm in S5 specifically includes the following steps:

s5.1, changing x to [ I_A,I_B,I_C,I₁,I₂,……I_N]Given to the input display layer, the probability P (h) of each hidden layer neuron h being activated is calculated by the following formula₁|v₁):

P(h_j|v)＝σ(b_j+∑_iW_i,jx_i)

In the formula, coefficient b_jBias for the jth apparent neuron, W_i,jThe weight value between any two connected neurons i and j is sigma function;

s5.2, extracting a sample h from the calculated probability distribution by adopting Gibbs sampling₁：

h₁～P(h₁|v₁)；

S5.3 use h₁Reconstructing the visualization layer, namely reversely deducing the visualization layer through the hidden layer, and calculating the probability P (v) that each visualization layer neuron v is activated by using the following formula₂|h₁)；

P(v_i|h)＝σ(c_i+∑_jW_i,jh_j)

In the formula, coefficient c_iA bias for the ith hidden layer neuron;

s5.4: similarly, a sample v is extracted from the calculated probability distribution using Gibbs sampling₂By v₂Calculating the activated probability of each neuron in the hidden layer again to obtain a probability score P (h)₂|v₂)：

v₂～P(v₂|h₁)；

S5.5: update the weight W with the contrast divergence method:

W←W+λ(P(h₁|v₁)v₁-P(h₂|v₂)v₂)

b←b+λ(v₁-v₂)

c←c+λ(h₁-h₂)

in the formula, lambda is the learning rate of the contrast divergence algorithm;

s6: carrying out supervised integral fine adjustment on the pre-trained mean square error function L (theta) of the DBN by adopting a gradient descent BP algorithm; after the DBN is fully trained, an intelligent control strategy model of the phase change switch of the transformer area to be verified is obtained;

s7: inputting the test sample set into a model to be verified, checking whether an error value U between a phase change switch state output value and an actual value is within a preset error range, if not, returning to S5, and continuing training to improve the model precision; and if so, obtaining a platform area phase change switch control model for making an on-line decision and controlling the platform area phase change switch.

2. The intelligent control method for the phase change switch used for three-phase unbalance treatment of the transformer area as claimed in claim 1, wherein the K-means algorithm specifically comprises the following steps:

s1.1: setting the clustering number of the commutation transient current curve as k, and initializing k clustering centers at random to obtain an initial clustering center; input N phase change switchesTaking the phase-change transient current curve of the load as sample data X_j，j＝1,2,…,N；

S1.2: calculating X_jDistance from the center of each cluster, X_jAnd (3) classifying the cluster centers with the closest distance into the corresponding classes, wherein the calculation formula is as follows:

dist(ω_i,X_j)＝||X_j-ω_i||²

in the formula, ω_iWhere i is 1,2, …, k denotes the ith cluster center, dist (ω)_i,X_j) Representing the distance between the clustering center and the sample data;

s1.3: computing X in each class_jThe calculation result is used as a new clustering center, and the calculation formula is as follows:

s1.4: and repeating S1.2 and S1.3 until the clustering center is not changed or the change is very small, so that the commutation switch is divided into k types according to the commutation transient current curve of the load connected with the commutation switch.

3. The intelligent control method for the phase change switches used for three-phase unbalance management of the transformer area as claimed in claim 2, wherein in the S1, the phase change switch with the minimum transient current amplitude is arranged at the front end of the sequence according to the ordering rule, and the intra-class phase change switches are also ordered in an ascending order according to the transient current amplitude.

4. The intelligent control method for the phase change switch used for the three-phase unbalance treatment of the transformer area according to claim 1, wherein the step of S2 making the control strategy of the phase change switch specifically comprises the following steps:

s2.1: initializing system parameters, and enabling t to be 0 and num to be 0, wherein t is a system timer, and num is the number of times of overrun of the three-phase current unbalance degree; setting a three-phase current monitoring period T, and setting a maximum allowable value delta and an allowable overrun number n of the three-phase current unbalance;

s2.2: when T is equal to T,calculating the average current I of the distribution area_aveAnd three-phase unbalance eta, and the calculation formula is as follows:

s2.3: judging whether eta exceeds delta or not, if not, resetting t to be 0, and returning to S2.2; if yes, executing S2.4;

s2.4: a, B, C calculating the difference delta I between the three-phase current and the average current_A、ΔI_B、ΔI_CThe calculation formula is as follows:

ΔI_A＝I_A-I_ave

ΔI_B＝I_B-I_ave

ΔI_C＝I_C-I_ave；

s2.5: controlling the commutation switch to carry out load adjustment according to the calculation result of the S2.4;

s2.6: judging whether the eta after phase conversion is lower than delta, if not, continuing to execute S2.5; otherwise, recording each phase current I of the phase conversion foreground area_A、I_B、I_CCurrent I corresponding to each phase change switch before phase change₁、I₂、……、I_NEach commutation switch action mark x_iForming a commutation switch control strategy sample, resetting t to 0, and returning to step S2.2.

5. The intelligent control method of the phase change switch for the three-phase unbalance management of the transformer area according to claim 4, wherein the specific adjustment method in S2.5 is as follows:

if I_A>I_B>I_CSelecting the switch with the connected load in the A phase from the class 1 commutation switch to the C phase, wherein the switching amount is

If the switchable quantity of the A-phase load in the class 1 phase change switch is insufficient, selecting the class 2 phase change switch to continue switching, and repeating the operation until the switchable quantity is not less than delta I;

if I_A>I_B＝I_CSelecting the switch with the connected load in the A phase from the class 1 commutation switch to the B phase, wherein the switching amount is

If the switchable quantity of the A-phase load in the class 1 phase change switch is insufficient, selecting the class 2 phase change switch to continue switching, and repeating the operation until the switchable quantity is not less than delta I;

if I_A＝I_B>I_CSelecting the switch with the connected load in the A phase from the class 1 commutation switch to the C phase, wherein the switching amount is

If the switchable quantity of the A-phase load in the class 1 phase change switch is insufficient, the class 2 phase change switch is selected to continue switching, and the like is repeated until the switchable quantity is not less than delta I.

6. The intelligent control method for the commutation switches used for the three-phase imbalance management of the transformer area as claimed in claim 4, wherein S2.6 action marks χ of each commutation switch_iE {1,2,3,4,5,6}, 1 denotes a switch from a phase to B phase, 2 denotes a switch from a phase to C phase, 3 denotes a switch from B phase to a phase, 4 denotes a switch from B phase to C phase, 5 denotes a switch from C phase to a phase, and 6 denotes a switch from C phase to B phase.

7. The intelligent control method for the commutation switch used for three-phase imbalance management in the transformer area of claim 1, wherein the ratio of the training sample set to the test sample set in S3 is 8: 2.

8. The commutation switch of claim 1, wherein the commutation switch is used for three-phase unbalance treatment of a transformer areaThe intelligent control method is characterized in that in S5, the number of input layer nodes of the DBN is equal to (N +3) of three-phase unbalanced electrical characteristic parameters of the transformer area, namely [ I [ ]_A,I_B,I_C,I₁,I₂,……I_N](ii) a The output layer corresponds to the action states of N phase change switches, and the number of nodes is N; the structure parameter of the initialized deep belief network is 3 hidden layers, and the number of the nodes of the initial hidden layers is s-2 m + 1.

9. The intelligent control method for the commutation switch used for three-phase imbalance management in the transformer area according to claim 1, wherein the supervised integral fine tuning is a top-down tuning of the pre-trained DBN, and a network parameter vector θ is updated as [ b, c, W ]:

U_i＝y_i-y_out

in the formula, alpha is the learning rate of BP algorithm, U is the error value, n_outThe number of output nodes; y is_iTo train the actual output value of the sample, y_outIs the predicted output value of the DBN.

Images