An automatic control and early warning method for a smart air switch device and its implementation platform

Abstract

The present invention relates to an automatic control and early warning method for a smart air switch device and its implementation platform, including: 1) using sensors to monitor and record the corresponding electrical and environmental parameters of the smart air switch device; 2) using feature engineering design methods to design three types of state features, and determining their mean and variance statistical characteristics according to the voltage, current, and temperature of the air switch device when it is opened and closed; 3) using a machine learning classification algorithm to use the multi-dimensional input features of the model after data processing, using the designed calibration method to divide the data into three labels, and divide the data into a training set, a validation set, and a test set, and perform learning training and test evaluation on the model; 4) designing a smart air switch monitoring and safety management platform to predict and judge the corresponding state and quickly execute related actions. The present invention can not only prevent serious safety accidents caused by failures of air switch devices, but also effectively improve the management efficiency of the power system and reduce complicated manual labor.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent air switches (miniature circuit breakers), and in particular to an automatic control and early warning method for an intelligent air switch device and an implementation platform thereof. Background Art

[0002] Air switches (abbreviated as "air switches") are widely used in substation cabinets. They can realize functions such as controlling circuit separation and connection and protecting electrical equipment. They are often used for the control operation of secondary circuits of primary equipment such as isolating switches.

[0003] At present, under the reform of the full sequential control mode of substations, in order to avoid the manpower and time consumption caused by the need for human cooperation in the operation of the power circuit breaker of the disconnecting switch (earthing switch) at the substation site, the power circuit breaker of the disconnecting switch (earthing switch) is usually preset to the closing position, so that the background can perform "one-click sequential control". However, this mode may cause the disconnecting switch (earthing switch) that is not within the working range to malfunction, and even cause a serious accident of mistakenly closing the grounding switch under power, which will "lay a mine" for the in-depth promotion of the full sequential control operation.

[0004] Under the current full control mode, the isolating switch power circuit breaker is usually placed in the closed position, which is convenient for background remote control, but also has great hidden dangers and inconvenience.

[0005] (I) May cause operational inconvenience

[0006] In full control mode, there are three situations in which the disconnector malfunctions due to the power supply of the disconnector being normally in the closed position.

[0007] (1) The cable insulation is low, causing the disconnector to malfunction.

[0008] (2) The insulation of the mechanism box is poor, causing the disconnector to malfunction. Due to water accumulation inside the mechanism box, short circuit of the terminal block due to moisture, short circuit of the terminal block due to foreign objects, AC ground loss, etc., if the main circuit power supply of the disconnector motor is not disconnected, once a certain pressure difference appears at the two ends A1 and A2 of the coils of the opening and closing relays KM1 and KM2, the contacts will be directly attracted, causing the main circuit of the motor to be turned on, causing the motor to rotate, driving the disconnector to open and close automatically, causing a vicious misoperation accident of the live disconnector.

[0009] (3) Human misoperation and accidental touch cause the disconnector to malfunction. Since the disconnector power supply is in the closed state, if someone touches the contactor by mistake, the disconnector will malfunction without considering the five protections of the microcomputer. For example, the contactor may be touched by mistake during the self-maintenance of the control cabinet.

[0010] Therefore, if a circuit breaker with automatic opening and closing capabilities can be provided, it can add a "firewall" to a certain extent to avoid the occurrence of misoperation accidents of the disconnector in the full control mode.

[0011] (II) Extending the time for power restoration

[0012] During the switching operation, the disconnector power supply circuit breaker needs to be frequently closed and disconnected, which takes a long time to execute the safety measures (it is necessary to open the doors of each compartment cabinet one by one and confirm the position status of each circuit breaker one by one.)

[0013] 3. Difficulty in achieving timely and accurate control

[0014] The fault judgment and maintenance of traditional air-breaker devices basically rely on instrument detection, experience judgment, regular maintenance and other methods. They can only be discovered when the fault of the air-breaker device occurs, but accidents are often difficult to avoid due to its lag. The automatic control and early warning method of a smart air-breaker device designed by the present invention makes full use of big data, artificial intelligence and Internet of Things technologies, uses the designed characteristic parameters as the input of the intelligent classification algorithm, and outputs the corresponding classification results to achieve accurate prediction, and realizes timely control through the remote control module of the cloud platform, solving a major pain point of traditional air-breaker devices.

[0015] Therefore, if an intelligent on-off circuit breaker can be provided, which can automatically pull out the power circuit breaker and confirm its position status after the equipment is in the correct state, it will inevitably greatly reduce the time for the installation arrangements required by the work ticket. Summary of the invention

[0016] In view of this, the purpose of the present invention is to provide an automatic control and early warning method for an intelligent air switch device and an implementation platform thereof, so that the intelligent air switch device and the cloud platform can be used to predict and judge the corresponding status and quickly execute related actions to avoid the occurrence of safety accidents. It can also effectively improve management efficiency and reduce complicated manual labor; the entire power operation and safety status of the factory or community can be managed and maintained.

[0017] The present invention is implemented by the following scheme: an automatic control and early warning method of a smart air switch device, comprising the following steps:

[0018] Step S1: using a sensor monitoring module to collect data including voltage, current, temperature and opening and closing speed of the isolating switch and the grounding switch of the air switch device during the opening and closing process; at the same time, collecting environmental parameters of the chassis where the air switch device is located, including temperature, humidity and abnormal noise;

[0019] Step S2: Using feature engineering methods, designing multi-dimensional statistical and state characteristics of air switch device parameters;

[0020] Step S3: Select a machine learning classification algorithm, train and deploy a model, and build a smart air switch device status classification algorithm model to monitor and warn of abnormal conditions of the air switch device and the chassis where it is located.

[0021] Furthermore, the step S1 specifically includes the following steps:

[0022] Step S11: Collect features: Collect circuit breaker device data including timestamp, product model type, rated current I e , voltage U e , and the current I when the disconnector is opened and closed t , voltage U t 、Ambient temperature T t , chassis temperature T tb 、Chassis humidity H tb , noise loudness N tb And the opening and closing speed v st ;

[0023] Step S12: Determine the rule: Determine the various electrical parameters collected from the multi-source sensors of the air switch device, including the rated current I e , voltage U e , and the current I when the disconnector is opened and closed t , voltage U t , environmental parameters include ambient temperature T t , chassis temperature T tb 、Chassis humidity H tb , noise loudness N tb And the opening and closing speed v st The mean, median, variance, and extreme values ​​of the data;

[0024] Step S13: Data selection: Draw a characteristic curve from the obtained statistics, i.e., the mean, median, variance and extreme value of the electrical parameters and environmental parameters determined in step S12, to obtain the basic law of the circuit breaker device during normal operation, and discard missing data and abnormal data;

[0025] If the collected data sample lacks one or more of the features collected in step S11, it is judged as missing data; if the features in S11 described in the collected sample have a value less than 0, it is judged as abnormal data and needs to be discarded.

[0026] Furthermore, the step S2 specifically includes the following steps:

[0027] Step S21: Design three status features according to short circuit, leakage and overload, including short circuit protection feature Per 1 , leakage protection feature Bolt and overload protection feature S t ;

[0028] When the short-circuit current value is reached, the main contact will be disconnected to cut off the circuit; the short-circuit current I reached within the preset time range is calculated by formula (1). th 90% of the total number of samples 1 :

[0029]

[0030] Among them, num(I th 0.9) represents the number of times that the short-circuit current threshold reaches 90% within this time range, and N represents the total number of samples collected within this time range;

[0031] Leakage protection features: When the circuit breaker is working normally, the current flowing into the circuit breaker device is equal to the current flowing out of the circuit breaker device. At this time, the induced voltage of the transformer is 0; if the inflow current is greater than the outflow current, the transformer generates an induced voltage, which is judged as leakage at the grounding point behind the circuit breaker device;

[0032] The state Bol at each moment is determined by formula (2) t :

[0033] Bol t =bool(comp[U st ,0]) (2)

[0034] Among them, U st It represents the induced voltage value of the transformer at a certain moment. The comp operation is used to compare its relationship with 0 and obtain its Boolean value through the bool operation: st =0, then Bol t =1, otherwise Bol t =0 indicates that there is leakage in the circuit breaker;

[0035] Overload protection feature: According to the standard, in the standard value of the circuit breaker product, taking into account the continuity of power supply and the overload capacity of the cable itself, the overload current is allowed to exist for a certain period of time, which is set to 1 hour. The state quantity S at each moment is determined by formula (3) t :

[0036]

[0037] Step S22: Calculate the statistical characteristics of the circuit breaker when it is opened and closed, including the current I t , voltage U t , Temperature T t The average and its standard deviation As shown in equations (4) to (9);

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044] Furthermore, the step S3 specifically includes the following steps:

[0045] Step S31: Select a multi-classification gradient boosting decision tree GBDT model to classify the circuit breaker into three types: normal, warning, and dangerous;

[0046] Step S32: classify the air switch device into three states, that is, the number of categories of the classification algorithm is 3;

[0047] The log-likelihood function of the true value y and the predicted value f(x) is expressed as shown in formula (10):

[0048]

[0049] Among them, if the sample output category is i, then y i =1, the probability of the i-th class p i (x) is expressed as formula (11)

[0050] As shown:

[0051]

[0052] Combining formula (10) and formula (11), the negative gradient error of the i-th sample in the p-th round corresponding to category k is obtained as:

[0053]

[0054] Afterwards, the best residual fitting value of each leaf node of the generated decision tree is approximately calculated by formula (13) to determine the output probability and error judgment of each type of result:

[0055]

[0056] Among them, err plk Represents the negative gradient error of the i-th sample corresponding to category k in the p-th round of training;

[0057] Step S33: For the GBDT classification algorithm, select the logarithmic loss function and calculate it using the above formula (10);

[0058] Step S34: Regularization is implemented by subsampling ratio method and step size definition method to prevent overfitting; wherein, the subsampling ratio should be selected appropriately, and here the ratio ratio is selected as 0.6, and 60% of the total number of samples are selected each time to fit the GBDT decision tree to reduce variance and prevent overfitting;

[0059] Define the step size step, whose value range is (0,1]. The smaller the step size, the more iterations are required. The step size and the maximum number of iterations together determine the fitting effect of the algorithm, as shown in formula (14):

[0060] f k (x) = f k-1 (x)+step*h k (x) (14)

[0061] Step S35: Model training and evaluation: The data of the air switch device obtained from step S1 and step S2 are used as the input features of the model; a data set is established for the data obtained from the multi-source sensors; the data set is divided into 80% training set and 20% test set, and 10% is extracted from the training set as the verification set; the training set and the verification set are labeled with values: normal, warning and dangerous;

[0062] The labeled training set data is put into the supervised training model, i.e., the gradient boosting decision tree GBDT classification algorithm described in steps S31 to S34, for learning and training; the model is evaluated by using the test set and the validation set and the precision and recall indicators, and the data is input into the model to obtain the corresponding prediction classification results, i.e., normal, warning and dangerous, so as to realize the monitoring and early warning of abnormal conditions of the air switch device and the chassis where it is located, i.e., warning or dangerous conditions, i.e., abnormal conditions;

[0063] The trained model is used to judge the real-time data transmitted from multi-source sensors. If it is judged to be a warning state, the self-check function is activated and the human administrator is reminded; if it is judged to be a dangerous state, the isolation switch is automatically disconnected immediately to avoid short circuit or leakage that endangers personal and property safety;

[0064] Use precision and recall to verify the performance of model training; the definition formula of each evaluation index is as follows:

[0065]

[0066]

[0067] Among them, TP means that the predicted result is positive and the actual result is also positive; TN means that the predicted result is negative and the actual result is also negative; FP means that the predicted result is positive and the actual result is negative; FN means that the predicted result is negative and the actual result is positive;

[0068] Collect fault data warnings or dangerous status. If the judgment result is true positive, the data can be retained in the cloud platform. If the type is false positive, adjust the fault data sample and model structure, and retrain to optimize the model to achieve high precision and recall. Use precision and recall evaluation indicators to evaluate the effect of model training.

[0069] Furthermore, the category label marking method is as follows: in the collected data samples of the circuit breaker device, if the circuit breaker device fails and cannot work normally, it is marked as dangerous; in the state characteristics designed in step S21, if one of the following three situations occurs, it is marked as a warning label;

[0070] That is, short circuit protection characteristic Per 1 Greater than 0.1, leakage protection characteristic Bol t Equal to 0 or overload protection characteristic S t =1 lasts longer than 1 hour or S t =2; other open data samples under normal working conditions are marked as normal.

[0071] Preferably, the present invention also provides an implementation platform of the automatic control and early warning method of a smart air switch device, including a data processing module, a feature design module, a decision-making module, a real-time display module and a remote control module; the data processing module performs the data processing described in step S12 and step S13 on the data collected from the multi-source sensors, calculates statistics and determines the rules, and discards missing data and abnormal data; the feature design module calculates the short-circuit protection feature Per according to the input feature quantity calculation method designed in step S21 and step S22 1 , leakage protection feature Bol t , Overload protection feature S t And the current I t , voltage U t , Temperature T t The average and its standard deviation Calculation is performed; the decision-making module deploys the GBDT classification model in step S3 to the cloud platform, and can judge and classify the input samples into three categories: normal, warning and dangerous; the real-time display module displays the characteristic values ​​of the smart air switch device in steps S11 to S13, the real-time and historical data of the parameters of the air switch device designed by S21 to S22, and the normal, warning or dangerous state of the smart air switch on the PC or mobile screen, so as to facilitate maintenance personnel to grasp the dynamic changes of the smart air switch system in time; the remote control module: if the smart air switch device is judged to be in a warning state, the self-check function is started, and information is sent to the real-time display module to remind the manual administrator; if it is judged to be in a dangerous state, the isolating switch is automatically disconnected immediately through remote control, and an alarm signal is triggered to prevent the air switch device from endangering personal and property safety due to short circuit or leakage.

[0072] Compared with the prior art, the present invention has the following beneficial effects:

[0073] (1) In addition to the traditional manual push rod and electric button control, the smart air switch of the present invention also uses the smart air switch monitoring and safety management platform to realize remote cloud control, which can realize local and remote integrated control solutions;

[0074] (2) The present invention makes full use of big data and Internet of Things technology, and adopts feature engineering methods to increase the dimension of input data, including the designed statistical and state quantity features, a total of 18 dimensions including the rated current I e , voltage U e , Current I when the isolating switch is open or closed t , voltage U t 、Ambient temperature T t , chassis temperature T tb 、Chassis humidity H tb , noise loudness N tb , opening and closing speed v st , short circuit protection characteristics Per 1 , leakage protection feature Bol t and overload protection feature S t 、Current I t , voltage U t , Temperature T t The average and its standard deviation To improve the accuracy and efficiency of model classification algorithm training;

[0075] (3) The present invention deploys and applies the trained model, and uses the intelligent circuit breaker device and cloud platform to predict and judge the corresponding status and quickly execute related actions to avoid the occurrence of safety accidents. It can also effectively improve management efficiency and reduce complicated manual labor; manage and maintain the entire power operation and safety status of the factory or community. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0077] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0078] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0079] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0080] like Figure 1 As shown, this embodiment provides an automatic control and early warning method for a smart air switch device, comprising the following steps:

[0081] Step S1: using a sensor monitoring module to collect data including voltage, current, temperature and opening and closing speed of the isolating switch and the grounding switch of the air switch device during the opening and closing process; at the same time, collecting environmental parameters of the chassis where the air switch device is located, including temperature, humidity and abnormal noise;

[0082] Step S2: Using feature engineering methods, designing multi-dimensional statistical and state characteristics of air switch device parameters;

[0083] Step S3: Select a machine learning classification algorithm, train and deploy a model, and build a smart air switch device status classification algorithm model to monitor and warn of abnormal conditions of the air switch device and the chassis where it is located.

[0084] In this embodiment, step S1 specifically includes the following steps:

[0085] Step S11: Collect features: Collect circuit breaker device data including timestamp, product model type, rated current I e , voltage U e , and the current I when the disconnector is opened and closed t , voltage U t 、Ambient temperature T t , chassis temperature T tb 、Chassis humidity H tb , noise loudness N tb And the opening and closing speed v st ;

[0086] Step S12: Determine the rule: Determine the various electrical parameters collected from the multi-source sensors of the air switch device, including the rated current I e , voltage U e , and the current I when the disconnector is opened and closed t , voltage U t , environmental parameters include ambient temperature T t , chassis temperature T tb 、Chassis humidity H tb , noise loudness N tb And the opening and closing speed v st The mean, median, variance, and extreme values ​​of the data;

[0087] Step S13: Data selection: Draw a characteristic curve from the obtained statistics, i.e., the mean, median, variance and extreme value of the electrical parameters and environmental parameters determined in step S12, to obtain the basic law of the circuit breaker device during normal operation, and discard missing data and abnormal data;

[0088] If the collected data sample lacks one or more of the features collected in step S11, it is judged as missing data; if the features in S11 described in the collected sample have a value less than 0, it is judged as abnormal data and needs to be discarded.

[0089] In this embodiment, step S2 specifically includes the following steps:

[0090] Step S21: Design three status features according to short circuit, leakage and overload, including short circuit protection feature Per 1 , leakage protection feature Bol t and overload protection feature S t ;

[0091] When the short-circuit current value is reached, the main contact will be disconnected to cut off the circuit; the short-circuit current I is calculated by formula (1) within the preset time range (here 10 days is selected as the set time range). th90% of the total number of samples 1 :

[0092]

[0093] Among them, num(I th 0.9) represents the number of times that the short-circuit current threshold reaches 90% within this time range, and N represents the total number of samples collected within this time range;

[0094] Leakage protection features: When the circuit breaker is working normally, the current flowing into the circuit breaker device is equal to the current flowing out of the circuit breaker device. At this time, the induced voltage of the transformer is 0; if the inflow current is greater than the outflow current, the transformer generates an induced voltage, which is judged as leakage at the grounding point behind the circuit breaker device;

[0095] The state Bol at each moment is determined by formula (2) t :

[0096] Bol t =bool(comp[U st ,0]) (2)

[0097] Among them, U st It represents the induced voltage value of the transformer at a certain moment. The comp operation is used to compare its relationship with 0 and obtain its Boolean value through the bool operation: st =0, then Bol t =1, otherwise Bol t =0 indicates that there is leakage in the circuit breaker;

[0098] Overload protection feature: According to the standard, in the standard value of the circuit breaker product, taking into account the continuity of power supply and the overload capacity of the cable itself, the overload current is allowed to exist for a certain period of time, which is set to 1 hour. The state quantity S at each moment is determined by formula (3) t :

[0099]

[0100] Step S22: Calculate the statistical characteristics of the circuit breaker when it is opened and closed, including the current I t , voltage U t , Temperature T t The average and its standard deviation As shown in equations (4) to (9);

[0101]

[0102]

[0103]

[0104]

[0105]

[0106]

[0107] Using feature engineering methods, we design input features for the collected data, and use the advantages of big data to solve the problem of poor model training results due to low data dimensions. Input features include rated current I e , voltage U e , Current I when the isolating switch is open or closed t , voltage U t 、Ambient temperature T t , chassis temperature T tb 、Chassis humidity H tb , noise loudness N tb , opening and closing speed v st , short circuit protection characteristics Per 1 , leakage protection feature Bol t and overload protection feature S t 、Current I t , voltage U t , Temperature T t The average and its standard deviation There are 18 feature dimensions in total.

[0108] In this embodiment, step S3 specifically includes the following steps:

[0109] Step S31: Algorithm selection: In order to realize the danger warning function of the intelligent circuit breaker device (such as short circuit, leakage and other unexpected situations), a multi-classification gradient boosting decision tree GBDT model (Gradient Boosting Decision Tree) is selected to divide the circuit breaker device into three types: normal, warning and dangerous; it is used for prediction when the circuit breaker device detects abnormal situations.

[0110] Compared with other classification algorithms, this algorithm can achieve higher training accuracy with relatively less parameter adjustment time; the model has a wide range of applications, high flexibility, and strong robustness to outliers.

[0111] Step S32: The circuit breaker is divided into three states, that is, the number of categories of the classification algorithm is 3; the log-likelihood function of the true value y and the predicted value f(x) is expressed as shown in formula (10):

[0112]

[0113] Among them, if the sample output category is i, then y i =1, the probability of the i-th class p i (x) is expressed as shown in formula (11):

[0114]

[0115] Combining formula (10) and formula (11), the negative gradient error of the i-th sample in the p-th round corresponding to category k is obtained as:

[0116]

[0117] Afterwards, the best residual fitting value of each leaf node of the generated decision tree is approximately calculated by formula (13) to determine the output probability and error judgment of each type of result:

[0118]

[0119] Among them, err plk Represents the negative gradient error of the i-th sample corresponding to category k in the p-th round of training;

[0120] Step S33: For the GBDT classification algorithm, an exponential loss function and a logarithmic loss function can generally be selected. Here, the logarithmic loss function is selected and calculated using the above formula (6);

[0121] Step S34: Regularization: Considering the overfitting problem, this embodiment implements regularization through a subsampling ratio method and a step size definition method to prevent overfitting; wherein, the subsampling ratio (subsample) should be selected appropriately, and here the ratio ratio is selected as 0.6, and 60% of the total number of samples are selected each time to fit the GBDT decision tree to reduce variance and prevent overfitting;

[0122] Define the step size step, whose value range is (0,1]. The smaller the step size, the more iterations are required. The step size and the maximum number of iterations together determine the fitting effect of the algorithm, as shown in formula (14):

[0123] f k (x) = f k-1 (x)+step*h k (x) (14)

[0124] Step S35: Model training and evaluation: The data of the air switch device obtained from step S1 and step S2 are used as the input features of the model; a data set is established for the data obtained from the multi-source sensors; the data set is divided into 80% training set and 20% test set, and 10% is extracted from the training set as the verification set; the training set and the verification set are labeled with values: normal, warning and dangerous;

[0125] The labeled training set data is put into the supervised training model, i.e., the gradient boosting decision tree GBDT classification algorithm described in steps S31 to S34, for learning and training; the model is evaluated by using the test set and the validation set and the precision and recall indicators, and the data is input into the model to obtain the corresponding prediction classification results, i.e., normal, warning and dangerous, so as to realize the monitoring and early warning of abnormal conditions of the air switch device and the chassis where it is located, i.e., warning or dangerous conditions, i.e., abnormal conditions;

[0126] The trained model is used to judge the real-time data transmitted from multi-source sensors. If it is judged to be a warning state, the self-check function is activated and the human administrator is reminded; if it is judged to be a dangerous state, the isolation switch is automatically disconnected immediately to avoid short circuit or leakage that endangers personal and property safety;

[0127] Use precision and recall to verify the performance of model training; the definition formula of each evaluation index is as follows:

[0128]

[0129]

[0130] Among them, TP means that the predicted result is positive and the actual result is also positive; TN means that the predicted result is negative and the actual result is also negative; FP means that the predicted result is positive and the actual result is negative; FN means that the predicted result is negative and the actual result is positive;

[0131] Collect fault data warnings or dangerous status. If the judgment result is true positive, the data can be retained in the cloud platform. If the type is false positive, adjust the fault data sample and model structure, and retrain to optimize the model to achieve high precision and recall. Use precision and recall evaluation indicators to evaluate the effect of model training.

[0132] In this embodiment, the category label marking method is as follows: in the collected data samples of the air switch device, if the air switch device fails and cannot work normally, it is marked as dangerous; in the state characteristics designed in step S21, if one of the following three situations occurs, it is marked as a warning label;

[0133] That is, short circuit protection characteristic Per 1 Greater than 0.1, leakage protection characteristic Bol t Equal to 0 or overload protection characteristic S t =1 lasts longer than 1 hour or S t =2; other open data samples under normal working conditions are marked as normal.

[0134] Preferably, the present embodiment also provides an implementation platform of the automatic control and early warning method of a smart air switch device, including a data processing module, a feature design module, a decision-making module, a real-time display module and a remote control module; the data processing module performs the data processing described in steps S12 and S13 on the data collected from the multi-source sensors, calculates statistics and determines the rules, and discards missing data and abnormal data; the feature design module calculates the short-circuit protection feature Per according to the input feature quantity calculation method designed in steps S21 and S22 1 , leakage protection feature Bol t , Overload protection feature S t And the current I t , voltage U t , Temperature T t The average and its standard deviation Calculation is performed; the decision-making module deploys the GBDT classification model in step S3 to the cloud platform, and can judge and classify the input samples into three categories: normal, warning and dangerous; the real-time display module displays the characteristic values ​​of the smart air switch device in steps S11 to S13, the real-time and historical data of the parameters of the air switch device designed by S21 to S22, and the state of the smart air switch, normal, warning or "dangerous, on the PC or mobile screen, so that maintenance personnel can grasp the dynamic changes of the smart air switch system in time. The module also includes an alarm function in a dangerous state; the remote control module: if the smart air switch device is judged to be in a warning state, the self-check function is started, and information is sent to the real-time display module to remind the manual administrator; if it is judged to be in a dangerous state, the isolation switch is automatically disconnected immediately by remote control, and an alarm signal is triggered to avoid the air switch device from endangering personal and property safety due to short circuit or leakage. Preferably, the specific implementation process of the platform in this embodiment is as follows:

[0135] Step 1: Deploy the trained model to the cloud system to monitor the status of data transmitted from the on-site environment in real time; use the cloud-based smart air switch management platform as the data middle platform to form a complete closed-loop structure of "data collection → feature engineering → model judgment → cloud processing → result feedback → execution control", which can realize the whole process of smart air switch devices from data monitoring to intelligent control.

[0136] Step 2: The judgment results are divided into three categories: normal, warning and dangerous. The results are fed back to the actuator (i.e., cloud control module) to decide whether to perform the corresponding opening and closing operations. The judgment classification results will also be displayed on the PC and mobile terminals for manual administrators to make maintenance decisions in a timely manner; the real-time data transmitted from multi-source sensors is judged using the model. If it is judged to be a warning state, the self-check function is started and the manual administrator is reminded; if it is judged to be a dangerous state, the isolation switch is automatically disconnected immediately to avoid endangering personal and property safety due to short circuit or leakage;

[0137] Step 3: Feedback the training results to the model in a timely manner, continuously optimize the model by adjusting the data and model parameters, and finally form a closed-loop control structure. Use smart circuit breakers and smart circuit breaker monitoring and safety management platform (cloud platform) to manage and maintain the entire power operation and safety status of the factory or community.

[0138] Preferably, in this embodiment, a well-designed algorithm model is integrated, and the Internet of Things technology is used to design an intelligent circuit breaker monitoring and safety management platform, i.e., a cloud platform; relevant data of the circuit breaker device is acquired in real time, parameter changes are monitored, and abnormal conditions are warned; when a danger warning occurs, the protection self-check function is started, and manual confirmation is performed; if a dangerous situation including a short circuit or leakage has occurred, the circuit is automatically disconnected in time to ensure the safety of people and property.

[0139] Preferably, in this embodiment, the state parameters and operation data of the smart air breaker device are obtained through the sensor detection device to realize the state monitoring of the air breaker device and the cabinet system in which it is located. The acquired historical data is dimensionally expanded by feature engineering, which is used as the input feature of the designed machine learning classification algorithm, and the model is trained and optimized to realize the dangerous state warning function of the smart air breaker and related power equipment. Taking advantage of the Internet of Things and big data technology, the algorithm model is integrated and deployed, and the smart air breaker monitoring and safety management platform is designed to realize the intelligent control and fault monitoring functions of the smart air breaker.

[0140] Preferably, in this embodiment, the air switch is used as the basic element, and the historical operation and current status data of the air switch and the chassis are obtained through multi-source sensors installed inside the air switch device and the distribution box, that is, the various electrical parameters and environmental data of the smart air switch are obtained, which includes a total of 6 real-time detection data, and are transmitted to the data cloud platform (air switch device management cloud platform) in real time to provide data support for the implementation of the fault prediction classification algorithm; the feature engineering method is used to perform feature design and dimension expansion on the acquired data, effectively improving the accuracy of the early warning classification model training; the gradient boosting decision tree (GBDT) is selected as the classification algorithm, and the processed data is input into the model for training; the model is deployed, and the Internet of Things technology is used to design a smart air switch monitoring and safety management platform to realize intelligent control and hazard warning functions, which can be used for industrial production and living property electricity safety management and control.

[0141] Preferably, in this embodiment, the smart circuit breaker includes three control modes: manual control of the circuit breaker push rod, automatic control by electric buttons, and intelligent cloud remote control, thus realizing an integrated control solution of manual, automatic and remote control.

[0142] The algorithm model designed in this embodiment combines the Internet of Things and big data technology to realize remote intelligent monitoring and control of smart circuit breakers. The status of the smart circuit breaker can be monitored in real time through historical data and current status, and it can be predicted and judged whether the circuit breaker, electrical equipment and the chassis are abnormal or faulty. If it is in the "warning" state, the function self-check is started and the maintenance personnel are reminded in time; if it is in the "dangerous" state, the opening and closing operations of the disconnector of the smart circuit breaker are remotely controlled.

[0143] Preferably, in this embodiment, after feature engineering, the designed multidimensional data features are used as input parameters of the classification algorithm and divided into three categories: training set, test set and validation set. The multi-classification gradient boosting decision tree (GBDT) is selected as the classification algorithm for the health status monitoring of the air switch device, and the model is trained and deployed.

[0144] After multiple iterations of training, the GBDT model is trained using test sets and validation sets to evaluate the training effect of the model. The classification results are output as "normal", "warning" and "dangerous", and the trained and optimized model is deployed to the air switch device management cloud platform. The real-time data transmitted from multi-source sensors is judged using the model. If it is judged to be in the "warning" state, the self-check function is started and the manual administrator is reminded; if it is judged to be in the "dangerous" state, the isolation switch is automatically disconnected immediately to avoid short circuits or leakages that endanger personal and property safety.

[0145] Collect fault data ("warning" or "dangerous" status). If the judgment result is "true positive", the data can be retained on the cloud platform; if the type is judged as "false positive", adjust the fault data sample and model structure, and retrain to optimize the model. Use evaluation indicators such as precision and recall rate to evaluate the effect of model training.

[0146] Preferably, in this embodiment, the cloud-based smart circuit breaker management platform is used as the data middle platform to form a complete closed-loop structure of "data collection → feature engineering → model judgment → cloud-based processing → result feedback → execution control", which can realize the whole process of smart circuit breaker devices from data monitoring to intelligent control.

[0147] Preferably, this embodiment obtains the electrical and environmental parameters of the smart air-breaker device and the electrical box where it is located through multi-source sensors, uses feature engineering methods to expand the data dimension, and forms multi-dimensional data features as input parameters of the classification judgment model; 3) selects a multi-classification gradient boosting decision tree (GBDT) as a status warning model for the smart air-breaker device; 4) uses regularization methods such as subsampling and step size setting to optimize the model to avoid overfitting problems and improve training accuracy and efficiency; 5) divides the input data into training sets, test sets, and validation sets, compares the test results with the output results after manual verification, and continuously adjusts the model and optimization parameters; 6) integrates and deploys the model, designs a cloud management platform for smart air-breaker devices, forms a closed-loop structure for management and control of smart air-breaker devices based on data-driven and Internet of Things technologies, and realizes an intelligent system that integrates the operation, management, maintenance, and control of related power devices.

[0148] The above description is only a preferred embodiment of the present invention. All equivalent changes and modifications made according to the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. An automatic control and early warning method for a smart air switch device, Features: The following steps are involved: Step S1: using a sensor monitoring module to collect data including voltage, current, temperature and opening and closing speed of the isolating switch and the grounding switch of the air switch device during the opening and closing process; at the same time, collecting environmental parameters of the chassis where the air switch device is located, including temperature, humidity and abnormal noise; Step S2: Using feature engineering methods, designing multi-dimensional statistical and state characteristics of air switch device parameters; Step S3: Select a machine learning classification algorithm, train and deploy a model, and build a smart air switch device status classification algorithm model to monitor and warn of abnormal conditions of the air switch device and its chassis; The step S2 specifically includes the following steps: Step S21: Design three status features according to short circuit, leakage and overload, including short circuit protection feature Per 1 , leakage protection feature Bol t and overload protection feature S t ; When the short-circuit current value is reached, the main contact will be disconnected to cut off the circuit; the short-circuit current threshold I is calculated within the preset time range by formula (1). th 90% of the total number of samples 1 : Among them, num(I th 0.9) represents the number of times that the short-circuit current threshold reaches 90% within this time range, and N represents the total number of samples collected within this time range; Leakage protection characteristics: When the circuit breaker is working normally, the current flowing into the circuit breaker device and the current flowing out of the circuit breaker device are equal. At this time, the induced voltage of the transformer is 0; if the inflow current is greater than the outflow current, the transformer generates an induced voltage, which is judged as leakage at the grounding point behind the circuit breaker device. The state Bol at each moment is determined by formula (2). t : Ball t =bool(comp[U st ,0]) (2) Among them, U st It represents the induced voltage value of the transformer at a certain moment. The comp operation is used to compare its relationship with 0 and obtain its Boolean value through the bool operation: st =0, then Bol t =1, otherwise Bol t =0 indicates that there is leakage in the circuit breaker; Overload protection feature: According to the standard, in the standard value of the circuit breaker product, taking into account the continuity of power supply and the overload capacity of the cable itself, the overload current is allowed to exist for a certain period of time, which is set to 1 hour. The state quantity S at each moment is determined by formula (3) t : Step S22: Calculate the statistical characteristics of the circuit breaker when it is opened and closed, including the current I t , voltage U t , Temperature T t The average and its standard deviation As shown in equations (4) to (9); The step S3 specifically comprises the following steps: Step S31: Use the GBDT model to classify the circuit breaker devices into three types: normal, warning and dangerous; Step S32: The circuit breaker is divided into three states, that is, the number of categories of the classification algorithm is 3; the log-likelihood function of the true value y and the predicted value f(x) is expressed as shown in formula (10): Among them, if the sample output category is k, then y k =1, the probability of the kth class p k (x) is expressed as shown in formula (11): Combining formula (10) and formula (11), the negative gradient error of the i-th sample in the p-th round corresponding to category k is obtained as: Afterwards, the best residual fitting value of each leaf node of the generated decision tree is approximately calculated by formula (13) to determine the output probability and error judgment of each type of result: Step S33: For the GBDT classification algorithm, select the logarithmic loss function and calculate it using the above formula (10); Step S34: Regularization is implemented by sub-sampling ratio method and step size definition method to prevent overfitting; wherein, the sub-sampling ratio should be selected appropriately, and here the ratio is selected as 0.6, and 60% of the total number of samples are selected each time to fit the GBDT decision tree; Define the step size step, whose value range is (0, 1]. The smaller the step size, the more iterations are required. The step size and the maximum number of iterations together determine the fitting effect of the algorithm, as shown in formula (14): f k' (x)=f k'-1 (x)+step*h k' (x) (14) Step S35: Model training and evaluation: The data of the air switch device obtained from step S1 and step S2 are used as the input features of the model; a data set is established for the data obtained from the multi-source sensors; the data set is divided into 80% training set and 20% test set, and 10% is extracted from the training set as the verification set; the training set and the verification set are labeled with values: normal, warning and dangerous; Put the labeled training set data into the supervised training model, i.e., the gradient boosting decision tree (GBDT) classification algorithm described in steps S31 to S34, for learning and training; evaluate the model using the precision and recall indicators of the test set and validation set, input the data into the model, and obtain the corresponding prediction classification results, i.e., normal, warning, and dangerous; The trained model is used to implement the judgment of the real-time data transmitted from multi-source sensors. If it is judged to be a warning state, the self-check function is activated and the human administrator is reminded; if it is judged to be a dangerous state, the isolation switch is automatically disconnected immediately; Use precision and recall to verify the performance of model training; the definition formula of each evaluation index is as follows: Among them, TP means that the predicted result is positive and the actual result is also positive; FP means that the predicted result is positive and the actual result is negative; FN means that the predicted result is negative and the actual result is positive; Collect fault data warnings or dangerous status. If the judgment result is true, the data will be retained in the cloud platform. If the type judgment is false positive, adjust the fault data sample and model structure, and retrain to optimize the model to achieve high precision and recall. Use precision and recall evaluation indicators to evaluate the effect of model training. The method of labeling the category label is as follows: in the collected data samples of the circuit breaker device, if the circuit breaker device fails and cannot work normally, it is marked as dangerous; in the state characteristics designed in step S21, if one of the following three situations occurs, it is marked as a warning label: That is, short circuit protection characteristic Per 1 Greater than 0.1, leakage protection characteristic Bol t Equal to 0 or overload protection characteristic S t =1 lasts longer than 1 hour or S t =2; other open data samples under normal working conditions are marked as normal.

2. According to claim 1, an automatic control and early warning method for a smart air switch device, Features: The step S1 specifically includes the following steps: Step S11: Collect features: Collect circuit breaker device data including timestamp, product model type, rated current I e , voltage U e , and the current I when the disconnector is opened and closed t , voltage U t 、Ambient temperature T t , chassis temperature T tb 、Chassis humidity H tb , noise loudness N tb And the opening and closing speed v st ; Step S12: Determine the rule: Determine the various electrical parameters collected from the multi-source sensors of the air switch device, including the rated current I e , voltage U e , and the current I when the disconnector is opened and closed t , voltage U t , environmental parameters include ambient temperature T t , chassis temperature T tb 、Chassis humidity H tb , noise loudness N tb And the opening and closing speed v st The mean, median, variance, and extreme values ​​of the data; Step S13: Data selection: Draw a characteristic curve from the obtained statistics, i.e., the mean, median, variance and extreme value of the electrical parameters and environmental parameters determined in step S12, to obtain the basic law of the circuit breaker device during normal operation, and discard missing data and abnormal data; If the collected data sample lacks one or more of the features collected in step S11, it is judged as missing data; if the features in S11 described in the collected sample have a value less than 0, it is judged as abnormal data and needs to be discarded.

3. According to claim 2, a device for implementing the automatic control and early warning method of a smart air switch device, It is characterized in that The invention comprises a data processing module, a feature design module, a decision-making module, a real-time display module and a remote control module; the data processing module performs the data processing described in step S12 and step S13 on the data collected from the multi-source sensors, calculates the statistics and determines the rules, and discards the missing data and abnormal data; the feature design module calculates the short-circuit protection feature Per according to the input feature quantity calculation method designed in step S21 and step S22 1 , leakage protection feature Bol t , Overload protection feature S t And the current I t , voltage U t , Temperature T t The average and its standard deviation Calculation is performed; the decision-making module deploys the GBDT classification model in step S3 to the cloud platform, and can judge and classify the input samples into three conditions: normal, warning and dangerous; the real-time display module displays the characteristic values ​​of the smart air-breaker device in steps S11 to S13, the real-time and historical data of the parameters of the air-breaker device designed by S21 to S22, and the normal, warning or dangerous state of the smart air-breaker on the PC or mobile screen, so as to facilitate maintenance personnel to grasp the dynamic changes of the smart air-breaker system in time; the remote control module: if the smart air-breaker device is judged to be in a warning state, the self-check function is started, and information is sent to the real-time display module to remind the manual administrator; if it is judged to be in a dangerous state, the isolating switch is automatically disconnected immediately through remote control, and an alarm signal is triggered.

Images