CN105467971B - A kind of second power equipment monitoring system and method - Google Patents

Abstract

The present invention proposes a secondary equipment monitoring system for electric power, including: a secondary equipment status monitoring server and at least one secondary equipment status monitoring terminal; wherein, the secondary equipment status monitoring server receives the monitoring information transmitted by the secondary equipment status monitoring terminal Data; the secondary equipment status monitoring terminal is used to monitor the status of the secondary equipment and send the obtained monitoring data to the secondary equipment status monitoring server. The invention also proposes a monitoring method for electric secondary equipment. The power secondary equipment monitoring system and method proposed by the invention improve the comprehensiveness and accuracy of secondary equipment fault risk diagnosis.

Description

A monitoring system and method for electric power secondary equipment

technical field

The invention relates to the field of power equipment monitoring, in particular to a power secondary equipment monitoring system and method.

Background technique

The safety of power equipment is the basis for the safe, stable and reliable operation of the power grid. Effective and accurate monitoring and diagnosis of power equipment is an effective way to improve the reliability of power supply and the intelligent level of power grid operation. With the rapid development of the power grid and power system to ultra-high voltage, large capacity, and large systems, the requirements for safety and reliability are getting higher and higher. The power system urgently needs more accurate and fast online monitoring and monitoring of power transmission and transformation equipment status Diagnosis technology, so data mining and information fusion technology will lead the new direction of power grid condition monitoring and be more and more widely used.

With the gradual opening of the power market, the competition in the power industry will increase, and condition-based maintenance has become a hot research topic in my country's power companies and scientific research institutions at all levels. Since my country's reform and opening up, the development of power electronics technology, computer technology and communication technology has been changing with each passing day, which has laid a solid foundation for the realization of condition-based maintenance of power equipment. At present, Chinese scholars have done more research on condition-based maintenance of primary equipment, but less research on condition-based maintenance of secondary equipment that implements protection, control, monitoring, and measurement functions for primary equipment.

For a long time, my country's secondary equipment has been using a time-based periodic maintenance system. For example, traditional relay protection equipment, according to relevant regulations and regulations, conduct regular inspections of relay protection equipment, safety automatic devices and secondary circuits. . The traditional regular maintenance method can ensure the function and normal operation of the secondary equipment to a certain extent, but there are also disadvantages. If the secondary equipment has defects or failures between two maintenances, it will have to wait until the next equipment maintenance Or it can only be discovered when the function of the device fails, and the abnormal operation status of the secondary equipment may cause significant losses to the power system. Therefore, there is an urgent need to evaluate the condition of secondary equipment, reasonably estimate the operating status of secondary equipment, implement condition-based maintenance, and effectively cooperate with the simultaneous development of condition-based maintenance of primary equipment to ensure the stable operation of the power system.

According to the difference in function and role, power equipment can be divided into primary equipment and secondary equipment, of which secondary equipment mainly includes relay protection equipment, safety equipment, automation equipment, DC equipment and communication equipment. The normal and reliable operation of the secondary equipment is directly related to the stable operation of the power system and the operation safety of the primary equipment. In the actual operation of the power grid, system accidents often occur due to the failure of secondary equipment. Taking relay protection as an example, with the development of the times and the advancement of technology, the correct action rate of relay protection shows an upward trend, but the number of incorrect actions is still high, which cannot be ignored. In the actual operation of relay protection, protection operators, design quality, manufacturing quality, natural disasters and operational errors may lead to protection malfunction or refusal to operate. With the popularization and application of microelectronics technology and computer technology, the operation reliability of new relay protection equipment has been greatly improved. The development promotes the transformation of the traditional periodic maintenance system to the direction of condition-based maintenance. The vigorous promotion of primary equipment condition maintenance will also drive the transformation of secondary equipment condition maintenance. Therefore, secondary equipment is in urgent need of in-depth research and expansion in terms of maintenance system, maintenance strategy, and maintenance cycle. It is imperative to realize the condition-based maintenance of secondary equipment, which is the need for the development of power systems.

Estimating the failure risk of secondary equipment based on the state parameters of the secondary equipment is a basic function and basic function in the condition-based maintenance technology of secondary equipment. How to improve the comprehensiveness and accuracy of failure risk diagnosis is an existing technology problems that need to be resolved.

Contents of the invention

To at least partially solve the problems existing in the prior art, the present invention proposes a power secondary equipment monitoring system, including: a secondary equipment status monitoring server and at least one secondary equipment status monitoring terminal; wherein,

The secondary equipment status monitoring server receives the monitoring data transmitted by the secondary equipment status monitoring terminal;

The secondary equipment status monitoring terminal is used to monitor the status of the secondary equipment and send the obtained monitoring data to the secondary equipment status monitoring server.

The power secondary equipment monitoring system, wherein,

The secondary equipment state monitoring server includes a machine learning model for fault diagnosis of secondary equipment, and the relationship between the fault risk value output by the power secondary equipment monitoring system and a given fault risk threshold according to the machine learning model, A fault risk level existing in the secondary equipment is judged.

The monitoring system for secondary electric equipment, wherein the given fault risk thresholds include threshold B, threshold C, and threshold D,

When the failure risk value is less than the threshold B, it indicates that the secondary equipment is operating normally;

When the failure risk value is greater than or equal to the threshold B and less than the threshold C, it indicates that the secondary equipment may be abnormal;

When the failure risk value is greater than or equal to the threshold C and less than the threshold D, it indicates that the secondary equipment has relatively serious defects;

When the failure risk value is greater than or equal to the threshold D, it indicates that the secondary equipment has serious defects.

In the monitoring system for secondary electric power equipment, the machine learning model is a support vector machine model.

In the power secondary equipment monitoring system, each type of secondary equipment uses its corresponding support vector machine model for fault diagnosis, and the input vector of the support vector machine model is the state parameter of the corresponding secondary equipment, that is, the secondary equipment The monitoring data obtained by the equipment status monitoring terminal.

The secondary equipment monitoring system for electric power, wherein, a support vector machine model is used for all secondary equipment to obtain the comprehensive failure risk level of all secondary equipment, and the following state parameters are used as input vectors of the support vector machine model: electronic Transformer sampling data quality parameters, merging unit sampling data quality parameters, merging unit power supply self-inspection information, substation main communication channel bit error rate, network switch receiving and sending data volume ratio, network message recording and analysis device record information completeness , self-inspection information of the hardware module of the relay protection device, CRC check code of the relay protection program, the communication transmission rate between the relay protection device and the process layer device, the completeness of the information sent by the relay protection device, and the temperature of the secondary equipment operating environment parameters, the humidity parameters of the secondary equipment operating environment, the correct rate of the feedback message sent by the intelligent terminal, the abnormality of the circuit breaker position indicator light, the working environment parameters of the uninterruptible power supply system, the load status of the uninterruptible power supply system, the Working hours, station AC power bus voltage status, substation important feeder line current status, DC bus and feeder insulation status, DC bus voltage offset degree, battery state of charge.

In the monitoring system for secondary electric power equipment, wherein the state parameters of the secondary equipment are weighted and then used as input vectors of the support vector machine model to participate in fault diagnosis, wherein the weight calculation process of the state parameters of the secondary equipment is as follows:

Step 1. Organize m experts to assign weights to n state parameters of secondary equipment, and each expert independently determines the weight value of n state parameters:

W _i1 , W _i2 ,...,W _ij ,...,W _in (1≤i≤m, 1≤j≤n,), where i represents the i-th expert, j represents the j-th state parameter , W _ij represents the weight value assigned by the i-th expert to the j-th state parameter;

Step 2. Calculate the average value of the weight values given by m experts:

Step 3. Obtain the deviation between the weight value and the weight average:

Step 4. W _ij whose deviation Δ _ij is greater than a given threshold needs to be reprocessed, and fed back to the i-th expert to redistribute the weight value of the j-th state parameter until all Δ _ij meet the requirements.

The present invention also proposes a monitoring method for electric secondary equipment, using the electric secondary equipment monitoring system to perform fault diagnosis on secondary equipment, including:

Step 100, the secondary equipment status monitoring terminal monitors the status of the secondary equipment, and sends the obtained monitoring data, that is, the status parameters of the secondary equipment, to the secondary equipment status monitoring server;

Step 200, the secondary equipment status monitoring server receives the monitoring data transmitted by the secondary equipment status monitoring terminal;

Step 300, the support vector machine model in the secondary equipment state monitoring server performs fault diagnosis on the secondary equipment according to the received state parameters of the secondary equipment.

The method for monitoring secondary electric power equipment also includes the process of training the support vector machine model before step 300:

First, the training set and test set are normalized in the same way, and the training set is used as the training sample of the support vector machine, and the support vector machine is trained by continuously optimizing the kernel function parameters. If the correct rate of the fault diagnosis result cannot reach If the requirements are met, it is necessary to reselect the parameter range of the kernel function until the correct rate of the diagnosis results meets the requirements. At this time, the support vector machine model that meets the requirements is obtained. Finally, the test set is used to verify the trained support vector machine. Whether the diagnosis is correct.

The method for monitoring secondary electric power equipment, wherein, using the support vector machine model to diagnose secondary equipment faults specifically includes:

(1) Obtain sample data of electrical secondary equipment with clear fault conclusions, divide the sample data of secondary equipment into training set and test set, and classify the sample data of secondary equipment according to the level of failure risk;

(2) Transform the secondary equipment sample data into a matrix, and use the same method to normalize the training set and test set respectively;

(3) To select an appropriate kernel function, first input a large data search range and use the grid search method to roughly select the parameter penalty factor c and kernel function δ, and then reasonably reduce the data search range on the basis of rough search, Precisely select the best parameters c and δ by grid search method;

(4) Utilize training set sample training to be based on the data model of support vector machine, and use test set sample to predict whether the diagnostic result meets the requirements, if not, then return to step (3) to reselect the parameter range of kernel function;

(5) Substitute the state parameter data of the secondary equipment that needs to be diagnosed into the model to obtain the diagnosis result.

The monitoring method for secondary electric power equipment, wherein the value range of the penalty factor c is set to [2 ^-10 , 2 ¹⁰ ] with a step of 0.4 by using the grid search method; the value range of the kernel function parameter δ is [ 2 ^-10 , 2 ¹⁰ ], the step is 0.4, through the training of the support vector machine, the optimal value of the penalty factor c is 0.83282, the optimal value of the kernel function parameter δ is 0.39227, and the selection parameter of the support vector machine classifier The accuracy rate is 77.5536%.

The monitoring method for secondary electric power equipment, wherein the value range of the penalty factor c is set to [2 ^-10 , 2 ⁰ ] with a step of 0.2 by using the grid search method; the value range of the kernel function parameter δ is [2 ^-10 , 2 ⁰ ], step 0.2, after training the support vector machine, the optimal value of the penalty factor c is 0.40421, the optimal value of the kernel function parameter δ is 1.00231, and the accuracy of the selection parameters of the support vector machine classifier is 93.1196%.

The monitoring method for secondary electric power equipment, wherein the value range of the penalty factor c is set to [2 ⁰ , 2 ¹⁰ ] with a step of 0.2 using the grid search method; the value range of the kernel function parameter δ is [2 ⁰ , 2 ¹⁰ ], step 0.2, after training the support vector machine, the optimal value of the penalty factor c is 1.2986, the optimal value of the kernel function parameter δ is 1.4093, and the accuracy rate of the parameter selection of the support vector machine classifier is 96.088%.

The monitoring method for secondary electric power equipment, wherein the value range of the penalty factor c is set to [2 ⁰ , 2 ¹⁰ ] with a step of 0.2, and the value range of the kernel function parameter δ is [2 ^{- 10} , 2 ⁰ ], step 0.2, through training the support vector machine, the optimal value of the penalty factor c is 23.2312, the optimal value of the kernel function parameter δ is 0.025102, and the accuracy rate of the parameter selection of the support vector machine classifier reaches 96.6598 %.

In the method for monitoring secondary electric power equipment, the support vector machine adopts a support vector machine based on particle swarm optimization, and the modeling process of the support vector machine based on particle swarm optimization is:

(1) Initialize the particle swarm, optimize the kernel function δ and the penalty factor c of the particle swarm support vector machine by adjusting the inertia weight of the particle swarm, so that the parameters c and δ constitute a particle, namely (c, δ), and Let the maximum velocity be V _max , use pbest to represent the initial position of each particle, and use gbest to represent the best initial position of all particles in the particle swarm;

(2) Evaluate the fitness of each particle and calculate the optimal position of each particle;

(3) Compare the fitness value of each particle after optimization with its historical best position pbest, if the current fitness value is better than the best position, take the fitness value as the particle’s current best position pbest;

(4) Compare the fitness value of each particle after optimization with the historical best position gbest of the population particle, if the fitness value is better than the historical best position gbest of the population particle, take the fitness value as the optimal position gbest of the population particle ;

(5) Adjust the velocity and position of the current particle according to the improved particle swarm optimization algorithm;

(6) When the fitness value satisfies the condition, the iteration ends, otherwise return to the second step to continue to optimize the parameters, and when the sixth step is completed, the best parameters c and δ will be optimized, so that the most ideal support vector can be obtained machine model, and use this model for fault prediction.

The monitoring method for secondary electric power equipment, wherein, set the population size N=20, the inertia weight ω=0.9, the acceleration constant C ₁ =1.4, C ₂ =1.6, and train the support vector machine to obtain the optimal selection of the penalty factor c The value is 3.8326, and the best value of kernel function δ is 0.50433. After optimization by particle swarm optimization, the classification accuracy of support vector machine classifier reaches 98.9234%.

The method for monitoring secondary electric power equipment, wherein the support vector machine adopts a support vector machine based on a genetic algorithm, and the modeling process of the support vector machine based on a genetic algorithm is:

(1) Initialize the population, generate a certain number of individuals as the initial population, each chromosome is composed of (c, δ), where c is the penalty factor, and δ is the kernel function;

(2) Select the objective function to carry out support vector machine training on the initial population, and use the mean square error of the support vector machine as the objective function to calculate the fitness of each individual;

(3) Perform selection operations, crossover operations, and mutation operations to obtain a new generation of populations, and perform support vector machine training on the newly generated populations;

(4) If the newly generated population satisfies the termination rule, output the individual with the maximum fitness as the optimal parameter, and use the optimal parameter for prediction, otherwise, increase the evolutionary number, and proceed to step (3) to continue.

A monitoring system and method for power secondary equipment proposed by the present invention can monitor the state parameters of power secondary equipment in real time, use machine learning models to predict and diagnose the failure risk of secondary equipment, and select appropriate state parameters through experiments As the input vector of the machine learning model, different weights are set for different state parameters taking into account the different effects on fault diagnosis, and various methods are used to improve the accuracy of machine learning model diagnosis.

Description of drawings

Fig. 1 is the schematic block diagram of the power secondary equipment monitoring system that the present invention proposes;

detailed description

The technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numerals in different drawings refer to the same or similar elements unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with aspects of the invention as recited in the appended claims.

Referring to Fig. 1, a monitoring system for power secondary equipment proposed by the present invention includes: a secondary equipment status monitoring server and at least one secondary equipment status monitoring terminal; wherein,

The secondary equipment status monitoring server receives the monitoring data transmitted by the secondary equipment status monitoring terminal;

The secondary equipment status monitoring terminal is used to monitor the status of the secondary equipment and send the obtained monitoring data to the secondary equipment status monitoring server.

In this embodiment, the secondary equipment state monitoring terminal performs real-time state detection on different secondary equipment, and sends the obtained state monitoring data to the secondary equipment state monitoring server. A secondary equipment status monitoring terminal can monitor only one secondary equipment, or monitor multiple secondary equipment at the same time.

Equipment status assessment mainly refers to the technical assessment of equipment status, based on comprehensive status information such as equipment operating conditions, load data, various status inspection data, defect information, fault and accident information, and maintenance data, according to regulations, operating experience, equipment manufacturers, etc. Criteria such as technical indicators, quantify and score the status information of the equipment, so as to judge and evaluate the real status of the equipment.

The present invention divides the secondary equipment status into four types:

A-Normal state: refers to the equipment with complete information, normal operation and various test data, and a slight deviation of individual data is allowed, as long as the trend of change is stable and there is no hidden danger in operation safety;

B-Suspicious state: refers to the equipment with unexplained defects or some test data showing that the equipment may be abnormal, but there are still some uncertain factors that cannot be determined;

C—reliability decline status: refers to equipment with relatively serious defects, or problems in the analysis of test results, and the location and cause of hidden dangers have been basically determined, and at the same time, the hidden dangers will not develop into accidents in a short period of time;

D-Dangerous state: refers to the serious defects of the equipment, or according to the test data, the operating conditions indicate that accidents may occur at any time.

Fault diagnosis is to judge the degree and cause of equipment abnormality through the abnormal phenomena displayed during equipment operation or maintenance.

There is no simple one-to-one correspondence between faults and symptoms, which makes fault diagnosis difficult. Due to the complexity of the relationship between equipment faults and symptoms and the complexity of equipment faults, equipment fault diagnosis is an exploratory method. Due to the characteristics of trial and error, the fault diagnosis process is complicated, and various mathematical diagnosis methods have their own advantages and disadvantages. The study of fault diagnosis methods has become the focus and difficulty of the discipline of equipment fault diagnosis technology, so a single method cannot be used for diagnosis. Instead, a variety of methods should be combined and applied in order to obtain the most correct diagnostic results that are closest to the facts, which is also the direction of the development of diagnostic methods in the future.

The secondary equipment status monitoring server in the present invention may include various machine learning models to perform fault diagnosis on the secondary equipment. According to the relationship between the failure risk value output by the machine learning model and a given failure risk threshold, the failure risk level of the secondary equipment is judged.

The given fault risk thresholds include threshold B, threshold C, and threshold D,

When the failure risk value is less than the threshold B, it indicates that the secondary equipment is operating normally;

When the failure risk value is greater than or equal to the threshold B and less than the threshold C, it indicates that the secondary equipment may be abnormal;

When the failure risk value is greater than or equal to the threshold C and less than the threshold D, it indicates that the secondary equipment has relatively serious defects;

When the failure risk value is greater than or equal to the threshold D, it indicates that the secondary equipment has serious defects.

An embodiment of the present invention adopts the support vector machine model to carry out fault diagnosis on the secondary equipment, and each kind of secondary equipment adopts its corresponding support vector machine model to carry out the fault diagnosis respectively, and the input vector of the support vector machine model is its corresponding secondary equipment The state parameters (that is, the monitoring data obtained by the secondary equipment state monitoring terminal), according to the relationship between the failure risk value output by the support vector machine model and the given failure risk threshold, determine the failure risk level of the secondary equipment .

Establishing the support vector machine model is the key and difficult point. Due to the large difference between the sample data, the sample data needs to be normalized. Before processing the sample, the sample data needs to be divided into two parts, one part is used as the training set, and the rest is used as the test set. The process of model training is as follows: firstly, the training set and test set are normalized in the same way, and the training set is used as the training sample of the support vector machine, and the support vector machine is trained by continuously optimizing the kernel function parameters. If the correct rate of the diagnostic result does not meet the requirements, the parameter range of the kernel function needs to be reselected until the correct rate of the diagnostic result meets the requirement. At this time, it is the most ideal support vector machine, and finally the training set is verified by the test set. Whether the fault diagnosis result of the vector machine is correct.

The specific implementation steps of support vector machine in secondary equipment fault diagnosis can be expressed as follows:

(1) Obtain sample data of electrical secondary equipment with clear fault conclusions, divide the sample data of secondary equipment into training set and test set, and classify the sample data of secondary equipment according to the level of failure risk;

(2) Transform the secondary equipment sample data into a matrix, and use the same method to normalize the training set and test set respectively;

(3) To select an appropriate kernel function, first input a large data search range and use the grid search method to roughly select the parameter penalty factor c and kernel function δ, and then reasonably reduce the data search range on the basis of rough search, Precisely select the best parameters c and δ by grid search method;

(4) Utilize training set sample training to be based on the data model of support vector machine, and use test set sample to predict whether the diagnostic result meets the requirements, if not, then return to step (3) to reselect the parameter range of kernel function;

(5) Substitute the state parameter data of the secondary equipment that needs to be diagnosed into the model to obtain the diagnosis result.

In order to make the support vector machine classifier achieve a higher classification accuracy and avoid the situation of "over-learning" and "under-learning" in the learning process, the cross-validation optimization support vector machine is selected, and the grid search method is used to select the best Optimal kernel function parameters. The principle is to divide the secondary equipment sample into two parts, take one part as the training set, and the remaining part as the test set. First, use the training set samples to train the support vector machine, use the grid search method to select the optimal parameters, construct a suitable decision function, and then use the test set samples to verify the trained support vector machine model to support the machine classification The accuracy rate of the fault judgment of the classifier is used as the performance index for evaluating the supporting machine classifier.

Sample data preprocessing

The present invention collects 400 sets of state parameter data for each secondary equipment as sample data, and each set of data has a definite fault conclusion. The secondary equipment sample data is divided into training set samples and test set samples, where the training set has 300 samples, and the remaining 100 samples are used as the test set. Convert the 400 sample data obtained by sorting into a matrix as the input data of the support vector machine,

Grid search method to select the best parameters c, δ

Import the normalized secondary equipment sample data into the database, and use the grid search method to select the optimal parameter δ and penalty factor c of the kernel function. Considering factors such as the data type and data volume of the secondary equipment sample comprehensively, the 10-fold cross-validation method is selected. Divide 400 secondary equipment sample data into 10 groups, take 8 of them and combine them as training sets, and the rest as test sets. After training, 10 times of classification accuracy will be obtained, and finally the arithmetic mean of 10 times of classification accuracy will be taken Values are used as performance metrics for support vector machine classifiers.

The grid search method sets the value range of the penalty factor c to [2 ^-10 , 2 ¹⁰ ] with a step of 0.4; the value range of the kernel function parameter δ is [2 ^-10 , 2 ¹⁰ ] with a step of 0.4. Through the training of the support vector machine, the optimal value of the penalty factor c is 0.83282, the optimal value of the kernel function parameter δ is 0.39227, and the accuracy rate of the parameter selection of the support vector machine classifier is 77.5536%.

Narrow the search range of the grid search method, and continue to train the support vector machine to find the optimal parameters and improve the accuracy of the support vector machine classifier selection parameters. The grid search method sets the value range of the penalty factor c as [2 ^-10 , 2 ⁰ ] with a step of 0.2; the value range of the kernel function parameter δ is [2 ^-10 , 2 ⁰ ] with a step of 0.2. After training the support vector machine, the optimal value of the penalty factor c is 0.40421, the optimal value of the kernel function parameter δ is 1.00231, and the accuracy rate of the parameter selection of the support vector machine classifier is 93.1196%.

The grid search method sets the value range of the penalty factor c to [2 ⁰ , 2 ¹⁰ ] with a step of 0.2; the value range of the kernel function parameter δ is [2 ⁰ , 2 ¹⁰ ] with a step of 0.2. After training the support vector machine, the optimal value of the penalty factor c is 1.2986, the optimal value of the kernel function parameter δ is 1.4093, and the accuracy rate of the parameter selection of the support vector machine classifier is 96.088%.

In order to analyze the impact of the penalty factor c and the kernel function parameter δ on the training of the support vector machine classifier, change the search range and continue training the support vector machine. The grid search method sets the value range of the penalty factor c as [2 ⁰ , 2 ¹⁰ ] with a step of 0.2, and the value range of the kernel function parameter δ as [2 ^-10 , 2 ⁰ ] with a step of 0.2. Through training the support vector machine, the optimal value of the penalty factor c is 23.2312, the optimal value of the kernel function parameter δ is 0.025102, and the accuracy rate of the parameter selection of the support vector machine classifier reaches 96.6598%.

It is not difficult to conclude from the above analysis that if the kernel function parameter δ value is too large or too small, it will cause "under-learning" or "over-learning" of the secondary equipment samples. The penalty factor c plays the role of adjusting the maximum classification interval and minimizing the training error. When the support vector machine classifier performs classification, if the value of the penalty factor c is large, the generalization ability of the support vector machine is poor; if the penalty factor c When the value is small, the generalization ability of the support vector machine is better. If the value of the penalty factor c exceeds a certain value, it will increase the complexity of the support vector machine and make it reach the maximum value required by the data space. Even if the range of the penalty factor c is enlarged, the training accuracy of the SVM will keep changing, but the testing accuracy of the SVM will no longer change.

The optimal value of penalty factor c is 23.2312 obtained by grid search method, the optimal value of kernel function parameter δ is 0.025102, and the accuracy rate of parameter selection of support vector machine classifier reaches 96.6598%. Using the qualified support vector machine classifier trained to predict the test set, input 100 secondary equipment samples in the test set to the support vector machine classifier, and the classification accuracy of the test set samples by the support vector machine reaches 93.36%.

The support vector machine model of the present invention can also adopt the support vector machine based on particle swarm optimization, and the modeling process of the support vector machine based on particle swarm optimization is:

(1) Initialize the particle swarm, optimize the kernel function δ and the penalty factor c of the particle swarm support vector machine by adjusting the inertia weight of the particle swarm, so that the parameters c and δ constitute a particle, namely (c, δ), and Let the maximum velocity be V _max , use pbest to represent the initial position of each particle, and use gbest to represent the best initial position of all particles in the particle swarm;

(2) Evaluate the fitness of each particle and calculate the optimal position of each particle;

(3) Compare the fitness value of each particle after optimization with its historical best position pbest, if the current fitness value is better than the best position, take the fitness value as the particle’s current best position pbest;

(4) Compare the fitness value of each particle after optimization with the historical best position gbest of the population particle, if the fitness value is better than the historical best position gbest of the population particle, take the fitness value as the optimal position gbest of the population particle ;

(5) Adjust the velocity and position of the current particle according to the improved particle swarm optimization algorithm;

(6) When the fitness value satisfies the condition, the iteration ends, otherwise return to the second step to continue to optimize the parameters, and when the sixth step is completed, the best parameters c and δ will be optimized, so that the most ideal support vector can be obtained machine model, and use this model for fault prediction.

Set population size N=20, inertia weight ω=0.9, acceleration constant C ₁ =1.4, C ₂ =1.6, train support vector machine, get the best value of penalty factor c 3.8326, the best value of kernel function δ is 0.50433. After particle swarm optimization, the classification accuracy of support vector machine classifier reaches 98.9234%.

The support vector machine model of the present invention can also be the support vector machine based on genetic algorithm, and the modeling process of the support vector machine based on genetic algorithm is:

(1) Initialize the population, generate a certain number of individuals as the initial population, each chromosome is composed of (c, δ), where c is the penalty factor, and δ is the kernel function;

(2) Select the objective function to carry out support vector machine training on the initial population, and use the mean square error of the support vector machine as the objective function to calculate the fitness of each individual;

(3) Perform selection operations, crossover operations, and mutation operations to obtain a new generation of populations, and perform support vector machine training on the newly generated populations;

(4) If the newly generated population satisfies the termination rule, then output the individual with the maximum fitness as the optimal parameter, and use the optimal parameter for prediction, otherwise increase the evolutionary algebra, and proceed to step (3) to continue to execute;

When the value of c obtained by the above method in the present invention is 50, and the value of δ is 0.52, the classification accuracy rate is 94.5%.

For different types of secondary equipment, the input vectors of the corresponding support vector machine models are different, that is, the state parameters of the input secondary equipment are different, but the training process and fault diagnosis process of the support vector machine are the same.

In the foregoing embodiments, different support vector machine models are used for different secondary devices, and one support vector machine model may be used for all secondary devices to obtain the comprehensive failure risk levels of all secondary devices. For the situation that all secondary devices adopt a support vector machine model, which state parameters are selected as the input vector of the support vector machine model needs comprehensive consideration and continuous testing. The present invention adopts the following state parameters as the input vector of the support vector machine model: electronic Transformer sampling data quality parameters, merging unit sampling data quality parameters, merging unit power supply self-inspection information, substation main communication channel bit error rate, network switch receiving and sending data volume ratio, network message recording and analysis device record information completeness , self-inspection information of the hardware module of the relay protection device, CRC check code of the relay protection program, the communication transmission rate between the relay protection device and the process layer device, the completeness of the information sent by the relay protection device, and the temperature of the secondary equipment operating environment parameters, the humidity parameters of the secondary equipment operating environment, the correct rate of the feedback message sent by the intelligent terminal, the abnormality of the circuit breaker position indicator light, the working environment parameters of the uninterruptible power supply system, the load status of the uninterruptible power supply system, the Working hours, station AC power bus voltage status, substation important feeder line current status, DC bus and feeder insulation status, DC bus voltage offset degree, battery state of charge.

Whether using different support vector machine models for different secondary equipment or using one support vector machine model for all secondary equipment, different state parameters of secondary equipment have different importance for the fault diagnosis of secondary equipment, so It is necessary to assign different weights to different state parameters, and then participate in fault diagnosis as the input vector of the support vector machine model.

The weight calculation process of the secondary equipment state parameters:

Step 1. Organize m experts to assign weights to n state parameters of secondary equipment, and each expert independently determines the weight value of n state parameters:

W _i1 , W _i2 ,...,W _ij ,...,W _in (1≤i≤m, 1≤j≤n,), where i represents the i-th expert, j represents the j-th state parameter , W _ij represents the weight value assigned by the i-th expert to the j-th state parameter;

Step 2. Calculate the average value of the weight values given by m experts:

Step 3. Obtain the deviation between the weight value and the weight average:

Step 4. W _ij whose deviation Δ _ij is greater than a given threshold needs to be reprocessed, and fed back to the i-th expert to redistribute the weight value of the j-th state parameter until all Δ _ij meet the requirements.

A monitoring system and method for power secondary equipment proposed by the present invention can monitor the state parameters of power secondary equipment in real time, use machine learning models to predict and diagnose the failure risk of secondary equipment, and select appropriate state parameters through experiments As the input vector of the machine learning model, different weights are set for different state parameters taking into account the different effects on fault diagnosis, and various methods are used to improve the accuracy of machine learning model diagnosis.

Other embodiments of the invention will be readily apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present invention, these modifications, uses or adaptations follow the general principles of the present invention and include common knowledge or conventional technical means in the technical field not disclosed in the present invention .

It should be understood that the present invention is not limited to the precise constructions which have been described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the invention is limited only by the appended claims.

Claims (5)

1. A power secondary equipment monitoring system, comprising: a secondary equipment status monitoring server and at least one secondary equipment status monitoring terminal; wherein, The secondary equipment status monitoring server receives the monitoring data transmitted by the secondary equipment status monitoring terminal; The secondary equipment status monitoring terminal is used to monitor the status of the secondary equipment and send the obtained monitoring data to the secondary equipment status monitoring server; in, The secondary equipment state monitoring server includes a machine learning model for fault diagnosis of secondary equipment, and the relationship between the fault risk value output by the power secondary equipment monitoring system and a given fault risk threshold according to the machine learning model, Judging the failure risk level of the secondary equipment; Wherein, the given fault risk threshold includes threshold B, threshold C, and threshold D, When the failure risk value is less than the threshold B, it indicates that the secondary equipment is operating normally; When the failure risk value is greater than or equal to the threshold B and less than the threshold C, it indicates that the secondary equipment may be abnormal; When the failure risk value is greater than or equal to the threshold C and less than the threshold D, it indicates that the secondary equipment has relatively serious defects; When the failure risk value is greater than or equal to the threshold D, it indicates that the secondary equipment has serious defects; Wherein, the machine learning model is a support vector machine model; Among them, each secondary equipment uses its corresponding support vector machine model for fault diagnosis, and the input vector of the support vector machine model is the state parameter of the corresponding secondary equipment, that is, the monitoring data obtained by the secondary equipment state monitoring terminal; Among them, the state parameters of the secondary equipment are given weights and then used as input vectors of the support vector machine model to participate in fault diagnosis. Among them, the weight calculation process of the state parameters of the secondary equipment: Step 1. Organize m experts to assign weights to n state parameters of secondary equipment, and each expert independently determines the weight value of n state parameters: W _i1 , W _i2 ,...,W _ij ,...,W _in , where, 1≤i≤m, 1≤j≤n, where i represents the i-th expert, and j represents the j-th state parameter , W _ij represents the weight value assigned by the i-th expert to the j-th state parameter; Step 2. Calculate the average value of the weight values given by m experts: <mrow><mover><msub><mi>W</mi><mrow><mi>i</mi><mi>j</mi></mrow></msub><mo>&amp;OverBar;</mo></mover><mo>=</mo><mfrac><mn>1</mn><mi>m</mi></mfrac><munderover><mo>&amp;Sigma;</mo><mrow><mi>i</mi><mo>=</mo><mn>1</mn></mrow><mi>m</mi></munderover><msub><mi>W</mi><mrow><mi>i</mi><mi>j</mi></mrow></msub><mo>;</mo></mrow> Step 3. Obtain the deviation between the weight value and the weight average: <mrow><msub><mi>&amp;Delta;</mi><mrow><mi>i</mi><mi>j</mi></mrow></msub><mo>=</mo><mo>|</mo><mo>|</mo><msub><mi>W</mi><mrow><mi>i</mi><mi>j</mi></mrow></msub><mo>-</mo><mover><msub><mi>W</mi><mrow><mi>i</mi><mi>j</mi></mrow></msub><mo>&amp;OverBar;</mo></mover><mo>|</mo><mo>|</mo><mo>;</mo></mrow> Step 4. W _ij whose deviation Δ _ij is greater than a given threshold needs to be reprocessed, and fed back to the i-th expert to redistribute the weight value of the j-th state parameter until all Δ _ij meet the requirements. 2. A method for monitoring power secondary equipment, using the power secondary equipment monitoring system as claimed in claim 1 to carry out fault diagnosis for secondary equipment, comprising: Step 100, the secondary equipment status monitoring terminal monitors the status of the secondary equipment, and sends the obtained monitoring data, that is, the status parameters of the secondary equipment, to the secondary equipment status monitoring server; Step 200, the secondary equipment status monitoring server receives the monitoring data transmitted by the secondary equipment status monitoring terminal; Step 300, the support vector machine model in the secondary equipment state monitoring server performs fault diagnosis on the secondary equipment according to the received state parameters of the secondary equipment. 3. The electric power secondary equipment monitoring method as claimed in claim 2, before step 300, also includes the process of training the support vector machine model: First, the training set and test set are normalized in the same way, and the training set is used as the training sample of the support vector machine, and the support vector machine is trained by continuously optimizing the kernel function parameters. If the correct rate of the fault diagnosis result cannot reach If the requirements are met, it is necessary to reselect the parameter range of the kernel function until the correct rate of the diagnosis results meets the requirements. At this time, the support vector machine model that meets the requirements is obtained. Finally, the test set is used to verify the trained support vector machine. Whether the diagnosis is correct. 4. The electric power secondary equipment monitoring method as claimed in claim 3, wherein, adopting the support vector machine model to carry out secondary equipment fault diagnosis specifically comprises: (1) Obtain sample data of electrical secondary equipment with clear fault conclusions, divide the sample data of secondary equipment into training set and test set, and classify the sample data of secondary equipment according to the level of failure risk; (2) Transform the secondary equipment sample data into a matrix, and use the same method to normalize the training set and test set respectively; (3) To select an appropriate kernel function, first input a large data search range and use the grid search method to roughly select the parameter penalty factor c and kernel function δ, and then reasonably reduce the data search range on the basis of rough search, Precisely select the best parameters c and δ by grid search method; (4) Utilize training set sample training to be based on the data model of support vector machine, and use test set sample to predict whether the diagnostic result meets the requirements, if not, then return to step (3) to reselect the parameter range of kernel function; (5) Substitute the state parameter data of the secondary equipment that needs to be diagnosed into the model to obtain the diagnosis result. 5. The monitoring method for power secondary equipment as claimed in claim 4, wherein, the value range of the penalty factor c set by the grid search method is [2 ⁰ , 2 ¹⁰ ], the step is 0.2, and the value of the kernel function parameter δ The value range is [2 ^-10 , 2 ⁰ ], step 0.2, through training the support vector machine, the optimal value of the penalty factor c is 23.2312, the optimal value of the kernel function parameter δ is 0.025102, the support vector machine classifier selection The accuracy of the parameters reaches 96.6598%.