Dry-type reactor defect online monitoring system and method

Abstract

The invention relates to an online defect monitoring system and method for a dry-type reactor, and the method comprises the steps: collecting a discharge signal through a signal collection module, and constructing a key information sample data set; the signal analysis module acquires key information to form an optimized sample data set, and performs model training and optimization on the sample data set based on a machine learning classification algorithm; and the defect prediction module carries out defect prediction on the optimization model and the collected real-time data, and pushes a maintenance strategy. According to the method, high-frequency current data are accumulated under different working conditions, various working conditions possibly existing in the reactor are learned through a machine learning classification algorithm, a detection model is trained, related parameters such as voltage, current and temperature are collected at the same time, a reactor defect model is optimized, defect prediction is carried out according to the high-frequency current data collected in real time, and the detection accuracy is improved. Therefore, the purposes of online monitoring of defects of the dry-type reactor and prevention of faults in advance are achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of reactor detection, and in particular to an online monitoring system and method for defects of a dry-type reactor. Background Art

[0002] With the development of UHV and EHV power systems in my country, dry-type reactors have been widely used in UHV and EHV power systems. In recent years, dry-type reactor accidents have occurred frequently, and fire accidents have also occurred from time to time, seriously threatening the safe operation of the power system. Reactor burnout, firefighting and power outages in accidents have caused huge economic losses to the system and also caused adverse social impacts. Therefore, it is necessary to promptly detect internal defects of dry-type reactors and quickly carry out inspection and maintenance operations.

[0003] There are two main types of defect detection for traditional reactors: online monitoring and power-off testing. Online monitoring is mainly based on infrared testing, which can effectively detect defects at metal joints, but it is difficult to detect defects inside the airway. Power-off testing mainly consists of DC resistance, inductance testing, turn-to-turn insulation testing, etc. Power-off testing is generally inefficient and can only detect serious defect turn-to-turn short-circuit faults, but cannot effectively detect short-circuits with a small number of turns. When a short-circuit fault with a small number of turns occurs during the operation of the reactor, it will often burn out quickly. If it is not removed in a short period of time, a power outage will occur. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and propose a system and method for online monitoring of dry-type inductor defects. By accumulating high-frequency current data under different working conditions, various possible working conditions of the inductor are learned through a machine learning classification algorithm and a detection model is trained. At the same time, relevant parameters such as voltage, current, and temperature are collected, the inductor defect model is optimized, and defect prediction is carried out based on the high-frequency current data collected in real time, thereby achieving the purpose of online monitoring of dry-type inductor defects and preventing faults from occurring in advance.

[0005] The present invention solves the technical problem by adopting the following technical solutions:

[0006] A dry-type reactor defect online monitoring system comprises a signal acquisition module, a signal analysis module and a defect prediction module, wherein the signal acquisition module, the signal analysis module and the defect prediction module are connected in sequence, the signal acquisition module is used for discharge signal acquisition, and at the same time, a key information sample data set is constructed; the signal analysis module forms an optimized sample data set by collecting key information and at the same time, a model training and optimization is performed on the sample data set based on a machine learning classification algorithm; the defect prediction module predicts defects by using the optimized model and the collected real-time data, and pushes a maintenance strategy.

[0007] Moreover, the signal acquisition module includes a single-core cable and a high-frequency CT. The single-core cable is installed at the neutral point of the reactor. The core of the single-core cable is suspended in the air. The cable casing is grounded through a long lead. The long lead is led outside the mesh door. The grounding wire does not contact the mesh door and a margin is left. The long lead is installed with a high-frequency CT for discharge signal acquisition.

[0008] Moreover, the high-frequency CT detects the high-frequency current of the neutral point of the reactor under different working conditions, including switching transient state and long-term operation steady state, obtains the high-frequency current pulse signal corresponding to the reactor under different working conditions, and constructs a key information sample data set.

[0009] A monitoring method for an online defect monitoring system for a dry-type reactor comprises the following steps:

[0010] Step 1: The signal acquisition module collects discharge signals and constructs a key information sample data set;

[0011] Step 2: The signal analysis module collects key information and forms an optimized sample data set, and performs model training and optimization on the sample data set based on a machine learning classification algorithm;

[0012] Step 3: The defect prediction module predicts defects through the optimization model and the collected real-time data, and pushes the maintenance strategy.

[0013] Moreover, the specific implementation method of step 1 is: install a single-core cable at the neutral point of the reactor, the core of the single-core cable is suspended in the air, the cable casing is grounded through a long lead, the long lead should be led outside the mesh door, the grounding wire does not contact the mesh door, and a margin is left, and a high-frequency CT is installed on the long lead for discharge signal collection; the high-frequency current of the neutral point of the reactor is detected under different working conditions by the high-frequency CT, including switching transient state and long-term steady state, to obtain the high-frequency current pulse signal corresponding to the reactor under different working conditions, and to construct a key information sample data set.

[0014] Moreover, the specific implementation method of forming the optimized sample data set in step 2 is: according to the reactor input and output status, the transient operating conditions of the reactor are divided into input conditions and output conditions; according to the continuous operation time, the steady-state operating conditions of the reactor are divided into three conditions: short-term input, medium-term input, and long-term input; the bus voltage, CT current, reactor infrared temperature measurement, encapsulated temperature sensor and discharge information obtained by high-frequency CT under transient and steady-state 5-minute conditions are collected, and historical test data including lightning impulse, reactance value and insulation value are also obtained to form an optimized sample data set, and modeling research is carried out respectively to construct the optimized sample data set:

[0015] Maximum, minimum, mean and variance of current;

[0016] Maximum, minimum, mean and variance of voltage;

[0017] Infrared temperature measurement time, temperature, and location;

[0018] Encapsulate the time, temperature and location of the temperature sensor;

[0019] The discharge volume obtained by high-frequency CT;

[0020] The switching status and time of the reactor in AVC.

[0021] Moreover, the specific implementation method of performing model training and optimization on the sample data set based on the machine learning classification algorithm in step 2 is: performing model training and optimization on the sample data set based on the machine learning classification algorithm, and the selected machine learning classification algorithm can adopt one or more algorithm fusions of commonly used algorithm models such as XGboost, lightGBM, random forest, support vector machine, naive Bayes, etc. for model training.

[0022] Moreover, the specific implementation method of step 3 is: using the established calculation model, respectively analyzing the characteristics of the reactor under five working conditions of input, output, short-term input, medium-term input and long-term input, and obtaining the intrinsic relationship between the degree of reactor defects and high-frequency discharge signals, voltage, current, historical test data and operating conditions, predicting the development of reactor defects through the trend of pattern data changes, and pushing different maintenance strategies for different defect degrees.

[0023] The advantages and positive effects of the present invention are:

[0024] The present invention collects discharge signals through a signal acquisition module and constructs a key information sample data set at the same time; the signal analysis module collects key information and forms an optimized sample data set, and at the same time performs model training and optimization on the sample data set based on a machine learning classification algorithm; the defect prediction module predicts defects through the optimized model and the collected real-time data, and pushes maintenance strategies. The present invention accumulates high-frequency current data under different working conditions, learns various working conditions that may exist in the reactor through a machine learning classification algorithm and trains a detection model, collects relevant parameters such as voltage, current, and temperature, optimizes the reactor defect model, and performs defect prediction based on the high-frequency current data collected in real time, thereby achieving the purpose of online monitoring of dry-type reactor defects and preventing faults in advance. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0026] The present invention is further described in detail below with reference to the accompanying drawings.

[0027] A dry-type reactor defect online monitoring system comprises a signal acquisition module, a signal analysis module and a defect prediction module, wherein the signal acquisition module, the signal analysis module and the defect prediction module are connected in sequence, the signal acquisition module is used for discharge signal acquisition, and at the same time, a key information sample data set is constructed; the signal analysis module forms an optimized sample data set by collecting key information and at the same time, a model training and optimization is performed on the sample data set based on a machine learning classification algorithm; the defect prediction module predicts defects by using the optimized model and the collected real-time data, and pushes a maintenance strategy.

[0028] The signal acquisition module includes a single-core cable and a high-frequency CT. The single-core cable is installed at the neutral point of the reactor. The core of the single-core cable is suspended in the air. The cable casing is grounded through a long lead. The long lead is led outside the mesh door. The grounding wire does not contact the mesh door and a margin is left. A high-frequency CT is installed on the long lead for discharge signal acquisition.

[0029] The high-frequency CT detects the high-frequency current at the neutral point of the reactor under different operating conditions, including switching transient and long-term steady-state operation, obtains the high-frequency current pulse signal corresponding to the reactor under different operating conditions, and constructs a key information sample data set.

[0030] A monitoring method for an online defect monitoring system of a dry-type reactor, such as Figure 1 As shown, the following steps are included:

[0031] Step 1: The signal acquisition module collects discharge signals and constructs a key information sample data set.

[0032] The specific implementation method of step 1 is as follows: install a single-core cable at the neutral point of the reactor, the core of the single-core cable is suspended in the air, and the cable casing is grounded through a long lead. The long lead should be led outside the grid door. The grounding wire does not touch the grid door and a margin is left. A high-frequency CT is installed on the long lead for discharge signal collection; the high-frequency current of the neutral point of the reactor is detected under different working conditions by the high-frequency CT, including switching transient state and long-term steady state operation, to obtain the high-frequency current pulse signal corresponding to the reactor under different working conditions, and to construct a key information sample data set.

[0033] Step 2: The signal analysis module collects key information and forms an optimized sample data set, and performs model training and optimization on the sample data set based on a machine learning classification algorithm.

[0034] The specific implementation method for forming the optimized sample data set in step 2 is: according to the reactor's input and output status, the transient operating conditions of the reactor are divided into input conditions and output conditions; according to the continuous operation time, the steady-state operating conditions of the reactor are divided into three conditions: short-term input (continuous operation less than 24 hours), medium-term input (continuous operation between 24 hours and 5068 hours), and long-term input (continuous operation greater than 5068 hours). Collect the bus voltage, CT current, infrared temperature measurement of the reactor, encapsulated temperature sensor (if any), and discharge information obtained by high-frequency CT under five working conditions such as transient and steady states. At the same time, historical test data including lightning impulse, reactance value, and insulation value are also obtained to form an optimized sample data set, and modeling studies are carried out separately to construct the optimized sample data set:

[0035] Maximum, minimum, mean and variance of current;

[0036] Maximum, minimum, mean and variance of voltage;

[0037] Infrared temperature measurement time, temperature, and location;

[0038] Encapsulate the time, temperature and location of the temperature sensor;

[0039] The discharge volume obtained by high-frequency CT;

[0040] The switching status and time of the reactor in AVC.

[0041] The specific implementation method of model training and optimization of the sample data set based on the machine learning classification algorithm in step 2 is: model training and optimization of the sample data set based on the machine learning classification algorithm. The selected machine learning classification algorithm can adopt one or more algorithm fusions of commonly used algorithm models such as XGboost, lightGBM, random forest, support vector machine, naive Bayes, etc. for model training.

[0042] Step 3: The defect prediction module predicts defects through the optimization model and the collected real-time data, and pushes the maintenance strategy.

[0043] The specific implementation method of step 3 is: using the established calculation model, analyze the characteristics of the reactor under five operating conditions: input, output, short-term input, medium-term input and long-term input, and obtain the intrinsic relationship between the degree of reactor defects and high-frequency discharge signals, voltage, current, historical test data and operating conditions. The development of reactor defects is predicted through the trend of pattern data changes, and different maintenance strategies are pushed for different defect degrees.

[0044] It should be emphasized that the embodiments described in the present invention are illustrative rather than restrictive. Therefore, the present invention includes but is not limited to the embodiments described in the specific implementation manner. Any other implementation manners derived by those skilled in the art based on the technical solution of the present invention also fall within the scope of protection of the present invention.

Claims

1. An online monitoring system for dry-type reactor defects, characterized in that: It includes a signal acquisition module, a signal analysis module and a defect prediction module, wherein the signal acquisition module, the signal analysis module and the defect prediction module are connected in sequence. The signal acquisition module is used for discharge signal acquisition and at the same time constructs a key information sample data set; the signal analysis module collects key information and forms an optimized sample data set, and at the same time performs model training and optimization on the sample data set based on a machine learning classification algorithm; the defect prediction module predicts defects through the optimized model and the collected real-time data, and pushes maintenance strategies.

2. The online defect monitoring system for dry-type reactor according to claim 1 is characterized in that: The signal acquisition module includes a single-core cable and a high-frequency CT. The single-core cable is installed at the neutral point of the reactor. The core of the single-core cable is suspended in the air. The cable casing is grounded through a long lead. The long lead is led outside the mesh door. The grounding wire does not contact the mesh door and a margin is left. The long lead is installed with a high-frequency CT for discharge signal acquisition.

3. The online defect monitoring system for dry-type reactor according to claim 2 is characterized in that: The high-frequency CT detects high-frequency current at the neutral point of the reactor under different working conditions, including switching transient state and long-term operation steady state, obtains high-frequency current pulse signals corresponding to the reactor under different working conditions, and constructs a key information sample data set.

4. A monitoring method for an online defect monitoring system for a dry-type reactor according to any one of claims 1 to 3, characterized in that: The following steps are involved: Step 1: The signal acquisition module collects discharge signals and constructs a key information sample data set; Step 2: The signal analysis module collects key information and forms an optimized sample data set, and performs model training and optimization on the sample data set based on a machine learning classification algorithm; Step 3: The defect prediction module predicts defects through the optimization model and the collected real-time data, and pushes the maintenance strategy.

5. A monitoring method for an online defect monitoring system for dry-type reactor according to claim 4, characterized in that: The specific implementation method of step 1 is: install a single-core cable at the neutral point of the reactor, the core of the single-core cable is suspended in the air, the cable casing is grounded through a long lead, the long lead should be led outside the mesh door, the grounding wire does not touch the mesh door, and a margin is left, and a high-frequency CT is installed on the long lead for discharge signal collection; the high-frequency current of the neutral point of the reactor is detected under different working conditions by the high-frequency CT, including switching transient state and long-term steady state, to obtain the high-frequency current pulse signal corresponding to the reactor under different working conditions, and to construct a key information sample data set.

6. A monitoring method for an online defect monitoring system for dry-type reactor according to claim 4, characterized in that: The specific implementation method of forming the optimized sample data set in step 2 is: according to the reactor input and output status, the transient operating conditions of the reactor are divided into input conditions and output conditions; according to the continuous operation time, the steady-state operating conditions of the reactor are divided into three conditions: short-term input, medium-term input, and long-term input; the bus voltage, CT current, reactor infrared temperature measurement, encapsulated temperature sensor and discharge information obtained by high-frequency CT under transient and steady-state conditions are collected, and historical test data including lightning impulse, reactance value and insulation value are also obtained to form an optimized sample data set, and modeling research is carried out respectively to construct the optimized sample data set: (1) Maximum, minimum, mean and variance of current; (2) Maximum, minimum, mean, and variance of voltage; (3) Infrared temperature measurement time, temperature and location; (4) Encapsulation time, temperature, and location of the temperature sensor; (5) discharge volume obtained by high-frequency CT; (6) The status and time of reactor switching on and off in AVC.

7. A monitoring method for an online defect monitoring system for dry-type reactor according to claim 4, characterized in that: The specific implementation method of the model training and optimization of the sample data set based on the machine learning classification algorithm in step 2 is: model training and optimization of the sample data set based on the machine learning classification algorithm. The selected machine learning classification algorithm can adopt one or more algorithm fusions of commonly used algorithm models such as XGboost, lightGBM, random forest, support vector machine, naive Bayes, etc. for model training.

8. The monitoring method for an online defect monitoring system for dry-type reactor according to claim 4, characterized in that: The specific implementation method of step 3 is: using the established calculation model, respectively analyzing the characteristics of the reactor under five operating conditions: input, output, short-term input, medium-term input and long-term input, and obtaining the intrinsic relationship between the degree of reactor defects and high-frequency discharge signals, voltage, current, historical test data and operating conditions, predicting the development of reactor defects through the trend of pattern data changes, and pushing different maintenance strategies for different defect degrees.

Images