Intelligent sensing network connection monitoring method and system for high-low voltage switch cabinet

Abstract

The invention discloses an intelligent sensing network connection monitoring method and system for a high-low voltage switch cabinet, and belongs to the technical field of electrical monitoring. The method comprises the following steps: acquiring physical quantity data of the high-low voltage switch cabinet according to an adaptive data acquisition period; generating a high-precision original data set through a signal processing algorithm and a data fusion algorithm; according to the high-precision original data set, the operation state of the high-low voltage switch cabinet is evaluated through a state evaluation algorithm, and a state evaluation result set is generated; and outputting a prediction result set of the high-low voltage switch cabinet through a state prediction method according to the state evaluation result set. According to the invention, on the basis of self-adaptive data acquisition, a state prediction method is adopted to accurately predict the fault occurrence probability of the high-low voltage switch cabinet according to the state evaluation result set, so that the effects of planning an operation and maintenance strategy in advance and ensuring stable operation of a power system are realized; the problem that the high-low voltage switch cabinet state prediction precision is low in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electrical monitoring, and in particular relates to an intelligent sensor network monitoring method and system for high and low voltage switch cabinets. Background Art

[0002] Sensor networking, or sensor network, is another name for the Internet of Things (IoT). With the rapid development of modern industry and the continuous acceleration of urbanization, the scale of the power system is becoming increasingly large and the structure is becoming increasingly complex. High and low voltage switchgear, as extremely critical equipment in the power system, bears important responsibilities such as distributing electric energy and controlling circuit switching. Under the high requirements for the continuous and stable operation of the power system, accurate monitoring of the operating status of high and low voltage switchgear has become crucial.

[0003] Traditional monitoring methods mostly rely on manual regular inspections and simple instrument readings. This method is not only inefficient, but also difficult to detect potential faults in a timely manner. The problem is often not known until the fault occurs, which leads to untimely maintenance, which may cause power outages and have serious impacts on industrial production and residents' lives. At the same time, some early automated monitoring methods also have limitations such as single functions and insufficient data accuracy, and cannot meet the needs of comprehensive, real-time and accurate monitoring of switch cabinets.

[0004] Based on the monitoring data, the future state of high and low voltage switchgear, especially potential faults, can be predicted. However, in the prior art, if the sampling frequency is set too high, a large amount of redundant data will be generated, which will increase the data processing burden and may introduce unnecessary noise; if the sampling frequency is too low, the key state change moment may be missed, and the dynamic changes of the switchgear operating state cannot be fully captured, resulting in the loss of important information during prediction and affecting the accuracy. In addition, the working conditions faced by high and low voltage switchgear in actual operation are extremely complex, such as different load levels, changes in ambient temperature and humidity, power grid fluctuations and other factors that affect each other and change dynamically. The existing prediction methods are difficult to fully consider the comprehensive impact of these complex working conditions on the state of the switchgear, and cannot dynamically adjust the prediction strategy according to the changes in working conditions in time, resulting in a large deviation between the prediction results and the actual state. Therefore, the prior art has the problem of low accuracy in the prediction of the state of high and low voltage switchgear. Summary of the invention

[0005] In view of the above-mentioned problems, the purpose of the present invention is to provide an intelligent sensor network monitoring method and system for high and low voltage switchgear. On the basis of adaptive data acquisition, a state prediction method is used to accurately predict the probability of failure of high and low voltage switchgear, thereby solving the problem of low accuracy of state prediction of high and low voltage switchgear in the prior art.

[0006] To achieve the above object, the present invention provides the following technical solutions: An intelligent sensor network monitoring method for high and low voltage switch cabinets comprises the following steps: Collect physical quantity data of high and low voltage switchgear according to the adaptive data collection cycle; Generate high-precision original data sets through signal processing algorithms and data fusion algorithms. The high-precision original data sets include real-time operating status data of the switch cabinet. Based on the high-precision original data set, the operating status of the high and low voltage switchgear is evaluated through the status evaluation algorithm to generate a status evaluation result set; According to the state assessment result set, the prediction result set of high and low voltage switchgear is output through the state prediction method.

[0007] Preferably, the analysis steps of the adaptive data collection cycle are: Physical quantity data include current, voltage, dielectric strength, and temperature; Calculate the current at time t ,Voltage , dielectric strength ,temperature The relative change includes: relative current change , relative voltage change , relative dielectric strength change , relative temperature change , the calculation formula is: ; Where t is the time index, For a fixed time interval, For in time The current value at For in time The voltage value at For in time The dielectric strength value at For in time Temperature value at ; Obtaining weight coefficients of physical quantity data for acquisition periods respectively; Get the benchmark collection period; The adaptive data collection period is obtained by combining the relative change amount, the weight coefficient and the benchmark collection period through the calculation formula of the adaptive data collection period.

[0008] Preferably, the calculation formula of the adaptive data collection period is: ; in, For the adaptive data collection cycle, is the benchmark collection period, , , , They are , , , Weight for the adaptive data collection period.

[0009] Preferably, the specific steps of the state assessment algorithm include: Collect historical operation data of high and low voltage switchgear from different regions, different service years and different models, and build a historical data set; After data cleaning and normalization of the historical data set, classification and labeling are performed, and hidden Markov models are constructed for different categories for training and learning, and the state transition probability matrix A, observation probability matrix B and initial state probability vector P are obtained; In real-time state evaluation, the newly collected real-time data is input into the corresponding hidden Markov model as an observation sequence, and the forward-backward algorithm is used to calculate the states of each state under the observation sequence O. Probability ; ; in, is the probability of the observation sequence O obtained by the forward algorithm, and They are the states corresponding to time t The forward and backward probabilities of By comparing the probabilities of each state and combining the preset state judgment threshold, the current operating state of the switch cabinet is determined.

[0010] Preferably, the specific steps of the state prediction method include: A feature selection method based on the fusion of information entropy and ReliefF algorithm is used to calculate the information entropy and importance score of each data in the state assessment result set, and the comprehensive importance weight is determined by weighted fusion. , select the data with the most value for prediction as the input of the LSTM model; The Xavier initialization method is used to initialize the weight matrix and bias vector of the long short-term memory network model, and then the long short-term memory network model is trained. The AdamW optimization algorithm is then used to optimize the trained long short-term memory network model, and the overfitting prevention strategy is started.

[0011] Preferably, the specific steps of the feature selection method include: Calculate the information entropy of each data , the information entropy calculation formula is: ; Among them, x represents different data types. Represents the kth value of data type x, The value of data x is probability; Use the ReliefF algorithm to calculate the importance score of each data ; The information entropy and importance score are normalized, and the comprehensive importance weight of each data is determined by weighted fusion. .

[0012] Preferably, the initialization formula for the weight matrix and bias vector of the initialization long short-term memory network model is: ; in, is the weight matrix, represents uniform distribution, is the number of input neurons, is the number of output neurons, is the bias vector.

[0013] Preferably, the specific optimization steps of the AdamW optimization algorithm include: Compute first- and second-order moment estimates: ; in, For the The first moment estimate of the round, For the Round second moment estimation, For the The gradient of the round, is the initial attenuation coefficient of the first-order moment estimate, is the initial attenuation coefficient of the second-order moment estimate, For the The first moment estimate of the round, For the Round second moment estimation; Modified first-order moment estimates and second-order moment estimates: ; in, After the correction The first moment estimate of the round, After the correction Round second moment estimation, To estimate the modified first-order moment The initial decay coefficient of the round, To estimate the modified second-order moment The initial decay coefficient of the round; According to the revised first-order moment estimate and second-order moment estimate, update the Round weight : ; in, Represents the trained long short-term memory network model in The weight of the round, is the initial learning rate, is a constant, is the weight decay coefficient.

[0014] Preferably, the specific steps of the strategy to prevent overfitting include: At the output end of the long short-term memory network model, the neuron connection is randomly discarded according to the set probability to perform the Dropout operation; When the validation set loss stops decreasing for Q consecutive rounds, training is stopped early and the EarlyStopping operation is performed.

[0015] An intelligent sensor network monitoring system for high and low voltage switchgear, used to implement the above method, includes: a data acquisition module, a data set generation module, a state evaluation module, and a state prediction module; Among them, the data acquisition module is used to collect physical quantity data of high and low voltage switch cabinets according to the adaptive data acquisition cycle; A data set generation module is used to generate a high-precision original data set through a signal processing algorithm and a data fusion algorithm. The high-precision original data set includes real-time operating status data of the switch cabinet; The state assessment module is used to assess the operating state of the high and low voltage switchgear based on the high-precision original data set and generate a state assessment result set through the state assessment algorithm; The state prediction module is used to output the prediction result set of the high and low voltage switchgear according to the state evaluation result set through the state prediction method.

[0016] The algorithm involved in the present invention can be executed by an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the above algorithm calculation is implemented by executing the software by the processor.

[0017] The beneficial effects of the present invention are: The present invention uses a state prediction method that integrates adaptive feature selection and optimized long short-term memory network models based on the state evaluation result set, thereby accurately predicting the probability of high and low voltage switchgear failures, thereby achieving the effect of planning operation and maintenance strategies in advance and ensuring the stable operation of the power system, and effectively solving the problem of low accuracy of high and low voltage switchgear state prediction in the prior art.

[0018] The present invention adopts an adaptive data collection cycle and dynamically adjusts the collection frequency according to the real-time operating status changes of the high and low voltage switchgear, thereby ensuring that the most valuable data can be obtained under different working conditions. It not only avoids the waste of resources caused by excessive data collection, but also prevents the loss of key information due to untimely collection, thereby achieving the technical effect of accurately capturing the details of the changes in the operating status of the switchgear.

[0019] The present invention uses a state evaluation algorithm based on a high-precision original data set to accurately determine the specific operating state of the high and low voltage switchgear, thereby achieving the effect of timely discovering potential fault hazards and improving the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of an intelligent sensor network monitoring method for high and low voltage switchgear in Example 1; Figure 2 This is a structural diagram of an intelligent sensor network monitoring system for high and low voltage switchgear in Example 2. DETAILED DESCRIPTION

[0021] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0022] Embodiment 1: like Figure 1 As shown, this embodiment provides an intelligent sensor network monitoring method for high and low voltage switch cabinets, comprising the following steps: collecting physical quantity data of the high and low voltage switch cabinets according to an adaptive data collection cycle; Generate high-precision original data sets through signal processing algorithms and data fusion algorithms. The high-precision original data sets include real-time operating status data of the switch cabinet. Based on the high-precision original data set, the operating status of the high and low voltage switchgear is evaluated through the status evaluation algorithm to generate a status evaluation result set; According to the state assessment result set, the prediction result set of high and low voltage switchgear is output through the state prediction method.

[0023] The signal processing algorithm can adopt the wavelet transform algorithm to decompose the collected signal into different frequency scales, remove noise interference and extract effective features; the data fusion algorithm can adopt the Kalman filter fusion algorithm to perform weighted fusion according to the error covariance matrix of each sensor data to improve the accuracy and reliability of the data.

[0024] The analysis steps of the adaptive data collection cycle are: Physical quantity data include current, voltage, dielectric strength, and temperature; Calculate the current at time t ,Voltage , dielectric strength ,temperature The relative change includes: relative current change , relative voltage change , relative dielectric strength change , relative temperature change , the calculation formula is: ; Where t is the time index, For a fixed time interval, For in time The current value at For in time The voltage value at For in time The dielectric strength value at For in time Temperature value at ; Obtaining weight coefficients of physical quantity data for acquisition periods respectively; Get the benchmark collection period; The adaptive data collection period is obtained by combining the relative change amount, the weight coefficient and the benchmark collection period through the calculation formula of the adaptive data collection period.

[0025] The calculation formula for the adaptive data collection period is: ; in, For the adaptive data collection cycle, is the benchmark collection period, , , , They are , , , Weight for the adaptive data collection period.

[0026] The weights are obtained from a database or set based on experience.

[0027] The specific steps of the state assessment algorithm include: Collect historical operation data of high and low voltage switchgear from different regions, different service years and different models, and build a historical data set; After data cleaning and normalization of the historical data set, classification and labeling are performed, and hidden Markov models are constructed for different categories for training and learning, and the state transition probability matrix A, observation probability matrix B and initial state probability vector P are obtained; In real-time state evaluation, the newly collected real-time data is input into the corresponding hidden Markov model as an observation sequence, and the forward-backward algorithm is used to calculate the states of each state under the observation sequence O. Probability ; ; in, is the probability of the observation sequence O obtained by the forward algorithm, and They are the states corresponding to time t The forward and backward probabilities of By comparing the probabilities of each state and combining the preset state judgment threshold, the current operating state of the switch cabinet is determined.

[0028] Set state set , is the last element. Each element in the state set corresponds to a different operating state, such as normal, minor fault, major fault, etc. The definition of the state is determined in combination with industry standards and actual operation and maintenance experience; observation sequence , including the corresponding real-time power, dielectric strength and other data collected.

[0029] The specific steps of the state prediction method include: A feature selection method based on the fusion of information entropy and ReliefF algorithm is used to calculate the information entropy and importance score of each data in the state assessment result set, and the comprehensive importance weight is determined by weighted fusion. , filter out the data with the most value for prediction as the input of the LSTM model.

[0030] The Xavier initialization method is used to initialize the weight matrix and bias vector of the long short-term memory network model, and then the long short-term memory network model is trained. The AdamW optimization algorithm is then used to optimize the trained long short-term memory network model, and the overfitting prevention strategy is started.

[0031] The specific steps of the feature selection method include: Calculate the information entropy of each data , the information entropy calculation formula is: ; Among them, x represents different data types. Represents the kth value of data type x, The value of data x is The probability of Use the ReliefF algorithm to calculate the importance score of each data ; The information entropy and importance score are normalized, and the comprehensive importance weight of each data is determined by weighted fusion. .

[0032] The specific process of using the ReliefF algorithm to calculate the importance score of each data is as follows: perform multiple random samplings in the sample space, extract a sample R each time, and then find its similar neighbor samples H (samples in the same operating state as R) and heterogeneous neighbor samples M (samples in different operating states from R), and update the importance score of each data according to the following formula: ; in, is the importance score of the updated data type x, , , Respectively represent the j-th eigenvalue of data type x in samples R, H, and M, and m is the number of samplings. After multiple iterative sampling updates, the importance score of each data based on the ReliefF algorithm is obtained.

[0033] Normalize the information entropy and importance score to get the normalized information entropy With the normalized importance , the normalized formulas are: ; ; in, , are the minimum and maximum values ​​of all data information entropy, respectively. , are the minimum and maximum values ​​of the importance scores of all data respectively. Then the weighted fusion method is used to determine the comprehensive importance weight of each data for prediction , let the information entropy weight be , the importance score weight is , then the rth data The calculation formula of the comprehensive importance weight is: ; in, For data The normalized information entropy of For data The normalized importance of .

[0034] The initialization formula for initializing the weight matrix and bias vector of the long short-term memory network model is: ; in, is the weight matrix, represents uniform distribution, is the number of input neurons, is the number of output neurons, is the bias vector.

[0035] The specific optimization steps of the AdamW optimization algorithm include: Compute first- and second-order moment estimates: ; in, For the The first moment estimate of the round, For the Round second moment estimation, For the The gradient of the round, is the initial attenuation coefficient of the first-order moment estimate, is the initial attenuation coefficient of the second-order moment estimate, For the The first moment estimate of the round, For the Round second moment estimation; Modified first-order moment estimates and second-order moment estimates: ; in, After the correction The first moment estimate of the round, After the correction Round second moment estimation, To estimate the modified first-order moment The initial decay coefficient of the round, To estimate the modified second-order moment The initial decay coefficient of the round; According to the revised first-order moment estimate and second-order moment estimate, update the Round weight : ; in, Represents the trained long short-term memory network model in The weight of the round, is the initial learning rate, is a constant, is the weight decay coefficient.

[0036] Usually take , , To prevent the denominator from being a small constant of zero, we usually take .

[0037] Specific steps to prevent overfitting strategies include: At the output end of the long short-term memory network model, the neuron connection is randomly discarded according to the set probability to perform the Dropout operation; When the validation set loss stops decreasing for Q consecutive rounds, training is stopped early and the EarlyStopping operation is performed.

[0038] Dropout operation: At the output end of the LSTM network model, neuron connections are randomly dropped with a set probability (such as 0.2). That is, for each neuron, there is a 0.2 probability that its output becomes 0 during each forward propagation process, so as to reduce the mutual dependence between neurons and increase the generalization ability of the model.

[0039] EarlyStopping operation: Divide the dataset into training set, validation set, and test set. After each round of training, use the validation set to evaluate the performance of the model, such as calculating the loss function value on the validation set. When the validation set loss stops decreasing for several consecutive rounds, stop training early and select the model parameters in the current state as the final model parameters to avoid overfitting of the model.

[0040] Embodiment 2: like Figure 2 As shown, an intelligent sensor network monitoring system for high and low voltage switch cabinets used to implement the method in Example 1 includes: a data acquisition module, a data set generation module, a state evaluation module, and a state prediction module; Among them, the data acquisition module is used to collect physical quantity data of high and low voltage switch cabinets according to the adaptive data acquisition cycle; A data set generation module is used to generate a high-precision original data set through a signal processing algorithm and a data fusion algorithm. The high-precision original data set includes real-time operating status data of the switch cabinet; The state assessment module is used to assess the operating state of the high and low voltage switchgear based on the high-precision original data set and generate a state assessment result set through the state assessment algorithm; The state prediction module is used to output the prediction result set of the high and low voltage switchgear according to the state evaluation result set through the state prediction method.

Claims

1. An intelligent sensor network monitoring method for high and low voltage switch cabinets, characterized in that: The following steps are involved: Collect physical quantity data of high and low voltage switchgear according to the adaptive data collection cycle; Generate high-precision original data sets through signal processing algorithms and data fusion algorithms. The high-precision original data sets include real-time operating status data of the switch cabinet. Based on the high-precision original data set, the operating status of the high and low voltage switchgear is evaluated through the status evaluation algorithm to generate a status evaluation result set; According to the state assessment result set, the prediction result set of high and low voltage switchgear is output through the state prediction method.

2. The intelligent sensor network monitoring method for high and low voltage switch cabinets according to claim 1 is characterized in that: The analysis steps of the adaptive data acquisition cycle are: Physical quantity data include current, voltage, dielectric strength, and temperature; Calculate the current at time t ,Voltage , dielectric strength ,temperature The relative change includes: relative current change , relative voltage change , relative dielectric strength change , relative temperature change , the calculation formula is: ； Where t is the time index, For a fixed time interval, For in time The current value at For in time The voltage value at For in time The dielectric strength value at For in time Temperature value at ; Obtaining weight coefficients of physical quantity data for acquisition periods respectively; Get the benchmark collection period; The adaptive data collection period is obtained by combining the relative change amount, the weight coefficient and the benchmark collection period through the calculation formula of the adaptive data collection period.

3. The intelligent sensor network monitoring method for high and low voltage switch cabinets according to claim 2 is characterized in that: The calculation formula of the adaptive data acquisition period is: ； in, For the adaptive data collection cycle, is the benchmark collection period, , , , They are , , , Weight for the adaptive data collection period.

4. The intelligent sensor network monitoring method for high and low voltage switch cabinets according to claim 1 is characterized in that: The specific steps of the state assessment algorithm include: Collect historical operation data of high and low voltage switchgear from different regions, different service years and different models, and build a historical data set; After data cleaning and normalization of the historical data set, classification and labeling are performed, and hidden Markov models are constructed for different categories for training and learning, and the state transition probability matrix A, observation probability matrix B and initial state probability vector P are obtained; In real-time state evaluation, the newly collected real-time data is input into the corresponding hidden Markov model as an observation sequence, and the forward-backward algorithm is used to calculate the states of each state under the observation sequence O. Probability ; ； in, is the probability of the observation sequence O obtained by the forward algorithm, and They are the states corresponding to time t The forward and backward probabilities of By comparing the probabilities of each state and combining the preset state judgment threshold, the current operating state of the switch cabinet is determined.

5. The intelligent sensor network monitoring method for high and low voltage switch cabinets according to claim 1 is characterized in that: The specific steps of the state prediction method include: A feature selection method based on the fusion of information entropy and ReliefF algorithm is used to calculate the information entropy and importance score of each data in the state assessment result set, and the comprehensive importance weight is determined by weighted fusion. , select the data with the most value for prediction as the input of the LSTM model; The Xavier initialization method is used to initialize the weight matrix and bias vector of the long short-term memory network model, and then the long short-term memory network model is trained. The AdamW optimization algorithm is then used to optimize the trained long short-term memory network model, and the overfitting prevention strategy is started.

6. The intelligent sensor network monitoring method for high and low voltage switch cabinets according to claim 5 is characterized in that: The specific steps of the feature selection method include: Calculate the information entropy of each data , the information entropy calculation formula is: ； Among them, x represents different data types. Indicates the kth value of data type x. The value of data x is probability; Use the ReliefF algorithm to calculate the importance score of each data ; The information entropy and importance score are normalized, and the comprehensive importance weight of each data is determined by weighted fusion. .

7. The intelligent sensor network monitoring method for high and low voltage switch cabinets according to claim 5 is characterized in that: The initialization formula for the weight matrix and bias vector of the initialization long short-term memory network model is: ； in, is the weight matrix, represents uniform distribution, is the number of input neurons, is the number of output neurons, is the bias vector.

8. The intelligent sensor network monitoring method for high and low voltage switch cabinets according to claim 7 is characterized in that: The specific optimization steps of the AdamW optimization algorithm include: Compute first- and second-order moment estimates: ； in, For the The first moment estimate of the round, For the Round second moment estimation, For the The gradient of the round, is the initial attenuation coefficient of the first-order moment estimate, is the initial attenuation coefficient of the second-order moment estimate, For the The first moment estimate of the round, For the Round second moment estimation; Modified first-order moment estimates and second-order moment estimates: ； in, After the correction The first moment estimate of the round, After the correction Round second moment estimation, To estimate the modified first-order moment The initial decay coefficient of the round, To estimate the modified second-order moment The initial decay coefficient of the round; According to the revised first-order moment estimate and second-order moment estimate, update the Round weight : ； in, Represents the trained long short-term memory network model in The weight of the round, is the initial learning rate, is a constant, is the weight decay coefficient.

9. The intelligent sensor network monitoring method for high and low voltage switch cabinets according to claim 7 is characterized in that: The specific steps of the strategy to prevent overfitting include: At the output end of the long short-term memory network model, the neuron connection is randomly discarded according to the set probability to perform the Dropout operation; When the validation set loss stops decreasing for Q consecutive rounds, training is stopped early and the EarlyStopping operation is performed.

10. An intelligent sensor network monitoring system for high and low voltage switch cabinets, used to implement the method described in any one of claims 1 to 9, characterized in that: include: Data acquisition module, data set generation module, state assessment module, state prediction module; Among them, the data acquisition module is used to collect physical quantity data of high and low voltage switch cabinets according to the adaptive data acquisition cycle; A data set generation module is used to generate a high-precision original data set through a signal processing algorithm and a data fusion algorithm. The high-precision original data set includes real-time operating status data of the switch cabinet; The state assessment module is used to assess the operating state of the high and low voltage switchgear based on the high-precision original data set and generate a state assessment result set through the state assessment algorithm; The state prediction module is used to output the prediction result set of the high and low voltage switchgear according to the state evaluation result set through the state prediction method.

Images