An intelligent monitoring system and method applied to a high-voltage switch cabinet
By optimizing the PD amplitude prediction of the LSTM model through phase judgment and data volume planning, the prediction deviation problem caused by cross-phase data changes is solved, and accurate fault identification and health management of high-voltage switchgear are realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGSU DEGANG POWER TECH CO LTD
- Filing Date
- 2026-03-06
- Publication Date
- 2026-06-09
AI Technical Summary
In existing technologies, LSTM models have difficulty capturing the cross-stage jump data changes in the amplitude of partial discharge (PD) in high-voltage switchgear, resulting in large deviations in prediction results and failing to effectively guarantee the accuracy of early identification of switchgear risks.
By determining the current stage of the PD amplitude of the high-voltage switchgear, data from multiple stages is removed and then input into the LSTM model. Combining recursive multi-step and direct multi-step prediction methods, the PD amplitude prediction of the LSTM model is optimized, the amount of data and the prediction duration are planned, and the impact of cross-stage data changes is reduced.
It improves the accuracy of PD amplitude prediction results, enhances the accuracy of early identification of switchgear risks, reduces the cumulative deviation of prediction results, and realizes early identification and health management of equipment failures.
Smart Images
Figure CN122173779A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of equipment monitoring technology, specifically an intelligent monitoring system and method for high-voltage switchgear. Background Technology
[0002] Partial discharge in high-voltage switchgear can cause local overheating and insulation damage. Monitoring the PD amplitude can detect whether insulation damage faults have occurred in the switchgear. However, simply monitoring the PD amplitude and issuing a fault warning when the PD amplitude reaches the alarm threshold may mean that the fault has already occurred and has affected the operational reliability and stability of the switchgear. Therefore, partial discharge prediction, i.e., PD amplitude prediction, for high-voltage switchgear can transform passive maintenance into proactive defense. PD amplitude prediction can help achieve early identification of switchgear risks and intervene before insulation degradation defects spread, avoiding power outages and safety risks caused by sudden faults such as breakdown accidents. However, in existing technologies, LSTM models are typically used to predict the PD amplitude at future moments. But the PD amplitude of switchgear generally exhibits changes in three stages: initial, middle, and late. Current technologies input all collected PD amplitudes into the LSTM model to predict the amplitude up to the point where a breakdown accident may occur. The input amplitudes may span multiple stages, and the amplitude change patterns differ across these stages. LSTM models themselves struggle to capture these cross-stage, jump-like data changes, easily leading to significant deviations in PD amplitude prediction results. Existing technologies have not optimized the method of using LSTM models for PD amplitude prediction, failing to effectively guarantee the accuracy of PD amplitude prediction results and thus hindering the accuracy of early risk identification in switchgear. Summary of the Invention
[0003] The purpose of this invention is to provide an intelligent monitoring system and method for high-voltage switchgear, so as to solve the problems raised in the prior art.
[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent monitoring method for high-voltage switchgear, the method comprising: S1: Collect historical PD amplitude data of the high-voltage switchgear that has been monitored and is to be predicted for PD amplitude, and determine the current stage of PD amplitude of different high-voltage switchgear. S2: Based on the current stage of the PD amplitude of the high-voltage switchgear, select the target historical PD amplitude data and input the selected target historical PD amplitude data into the LSTM model; S3: Collect the PD amplitude data of a randomly recorded high-voltage switchgear that has completed service, and make an initial estimate of the time required for the PD amplitude of different high-voltage switchgear to enter the next stage. S4: Based on the amount of historical PD amplitude data selected from the target data, plan the method of using the LSTM model to predict the PD amplitude. The method includes: Method 1: first perform recursive multi-step prediction, and then switch to direct multi-step prediction; Method 2: perform direct multi-step prediction. S5: Based on the initial estimated duration, plan the amount of data required for PD amplitude prediction using the LSTM model; S6: Based on the data output by the LSTM model, make a secondary prediction of the time required for the PD amplitude to enter the next stage, and provide equipment fault prompts based on the secondary prediction results.
[0005] Preferably, in step S1: the specific stage determination method is as follows: obtain the PD amplitude values of the two most recently monitored high-voltage switchgears as Q1 and Q2, set the PD amplitude monitoring interval time as t, calculate the PD amplitude growth rate W of the two most recently monitored high-voltage switchgears according to the formula W=(Q2-Q1) / t, and set the PD amplitude growth rate threshold intervals for the initial, middle and late stages as (0,w1), [w1,w2] and (w2,+∞), respectively, where w1 and w2 are the starting and ending values of the PD amplitude growth rate threshold interval for the middle stage, respectively. If W is within (0,w1), the current stage of the PD amplitude of the corresponding high-voltage switchgear is determined to be the initial stage; if W is within [w1,w2], the current stage of the PD amplitude of the corresponding high-voltage switchgear is determined to be the middle stage; if W is within (w2,+∞), the current stage of the PD amplitude of the corresponding high-voltage switchgear is determined to be the late stage.
[0006] Preferably, in step S2: if the PD amplitude of the high-voltage switchgear is currently in the initial stage, the target historical PD amplitude data to be input into the LSTM model is: all historical PD amplitude data of the corresponding high-voltage switchgear that have been monitored; if the PD amplitude of the high-voltage switchgear is currently in the intermediate stage, the target historical PD amplitude data to be input into the LSTM model is: all the remaining data after removing the PD amplitude data of the initial stage. If the PD amplitude of the high-voltage switchgear is currently in the later stage, the target historical PD amplitude data that needs to be input into the LSTM model is: all the remaining data after removing the PD amplitude data of the initial and intermediate stages. This invention addresses the issue that LSTM models struggle to capture data changes that jump across stages. It prioritizes determining the current stage of the PD amplitude of different high-voltage switchgear. Data containing multiple stages is removed from the monitored data before being input into the LSTM model for future PD amplitude prediction. The PD amplitude of each stage exhibits different data patterns, ensuring that the input data is from a single stage and reducing the probability of serious deviations in the LSTM model's prediction results due to the input of data from multiple stages.
[0007] Preferably, in step S3: the high-voltage switchgear that has completed service and the high-voltage switchgear to be predicted for PD amplitude are of the same type. For high-voltage switchgear whose PD amplitude is in the initial stage, the initial estimation method for the time required for the PD amplitude to enter the intermediate stage is as follows: obtain the PD amplitude growth rate R of the high-voltage switchgear whose PD amplitude is in the initial stage in the two most recent monitoring times; retrieve a random set of PD amplitudes of high-voltage switchgear that has completed service and is in the initial stage; arrange the PD amplitudes in the set in order from the time of acquisition to the time of acquisition; calculate the growth rate of every two adjacent PD amplitudes in the set; the calculation method is the same as the calculation method of W; and obtain the set of growth rates of every two adjacent PD amplitudes in the set as r={r1,r2,...,r m The set contains m+1 PD amplitudes, r m Let r represent the growth rate of the m-th and (m+1)-th PD amplitudes within the set. The growth rate of the m-th and (m+1)-th PD amplitudes is the rate at which the m-th PD amplitude changes to the (m+1)-th PD amplitude. Determine if there exists a value in set r equal to R. If only one value equal to R exists, the value equal to R is r. i Let T1 be the acquisition time of the i-th PD amplitude in the set, and T2 be the time when the PD amplitude of the corresponding high-voltage switchgear that has completed service enters the mid-term stage. Then, the initial estimated duration is H, where H = T2 - T1. If there is more than one value equal to R, the first initial estimated duration is obtained based on the first value in the set that is equal to R. If there is no value equal to R, the initial estimated duration is obtained based on the value in the set with the smallest difference from R.
[0008] Preferably, for high-voltage switchgear with PD amplitude in the middle stage: the initial estimation method for the time required for PD amplitude to enter the later stage is the same as the initial estimation method for the time required for PD amplitude to enter the middle stage for high-voltage switchgear with PD amplitude in the early stage; for high-voltage switchgear with PD amplitude in the later stage: the initial estimation method for the time required for insulation breakdown is the same as the initial estimation method for the time required for PD amplitude to enter the middle stage for high-voltage switchgear with PD amplitude in the early stage.
[0009] Preferably, in step S4: for a high-voltage switchgear cabinet whose PD amplitude is to be predicted, randomly select a high-voltage switchgear cabinet in the initial stage of the current PD amplitude stage: obtain a total of k target historical PD amplitude data to be input into the LSTM model, and obtain the average number L of target historical PD amplitude data to be input into the LSTM model for all high-voltage switchgear cabinets whose PD amplitudes are to be predicted. Compare k and L: if k < L, the method of using the LSTM model to predict the PD amplitude for the corresponding switchgear cabinet is Method 1; if , the method of using the LSTM model to predict the PD amplitude for the corresponding switchgear cabinet is Method 2.
[0010] Preferably, in step S5: for a high-voltage switchgear cabinet with the PD amplitude in the initial stage: plan the amount of data to be predicted B1 for predicting the PD amplitude using the LSTM model according to the formula B1 = H / t, that is, a total of B1 PD amplitudes need to be predicted, and perform a floor operation on B1. For high-voltage switchgear cabinets with the PD amplitude in the middle stage and the late stage, the planning method for the amount of data to be predicted is the same as that in the initial stage, and the amounts of data to be predicted for predicting the PD amplitude using the LSTM model for high-voltage switchgear cabinets with the PD amplitude in the middle stage and the late stage are obtained as B2 and B3 respectively; Based on the initial estimation duration and the PD amplitude monitoring interval time, the present invention plans the amount of data to be predicted for high-voltage switchgear cabinets in different stages, that is, how many PD amplitudes need to be predicted, which improves the possibility that the data output by the model does not have cross-stage data, increases the probability of using the LSTM model to only perform amplitude prediction in a single stage, further improves the problem that the LSTM model is difficult to capture cross-stage jump data changes and easily causes large deviations in the PD amplitude prediction results, optimizes the method of using the LSTM model to predict the PD amplitude, effectively ensures the accuracy of the PD amplitude prediction result, and further improves the accuracy of early risk identification of the switchgear cabinet.
[0011] Preferably, for a randomly selected high-voltage switchgear cabinet in the initial stage of the current PD amplitude stage: the operation method of Method 1 is: if k < L, set a non-arithmetic increasing sequence whose last term value is greater than or equal to L: the sequence is {a1, a2,..., a x}, where x represents the number of sequence terms, the initial value of the sequence a1 = k, the second term value of the sequence a2 = k + 1, a3 = a2 + a1, a x = a x-1 + a x-2The recursive multi-step prediction is performed based on the set sequence: First, one neuron is designed in the output layer of the LSTM model, with each neuron corresponding to a predicted value for one time step. The k historical target PD amplitudes are input into the LSTM model, which outputs the PD amplitudes for the next a2-a1 time steps. These future a2-a1 PD amplitudes are then added to the target historical PD amplitudes to generate the first new target historical PD amplitude. This new target historical PD amplitude is then input into the LSTM model, which outputs the PD amplitudes for the next a3-a2 time steps. These output values are then added to the first new target historical PD amplitude to generate the second new target historical PD amplitude. This second new target historical PD amplitude is then input into the LSTM model, which outputs the PD amplitudes for the next a4-a3 time steps. This process continues, with the output results used as input data to the model until the next a2-a3 time step is output. x -a x-1 The PD amplitude at each time step is calculated, and the output data is added to the input data to obtain the latest input data. Then, the method is switched to direct multi-step prediction: the latest input data is input into the LSTM model, and the LSTM model is used to directly output B1 PD amplitudes at once. For high-voltage switchgear where the current PD amplitude is in the middle and late stages, the operation method of Method 1 is the same as that in the initial stage.
[0012] Preferably, the operation method of the second method is as follows: for high-voltage switchgear where the PD amplitude is in the initial, middle and late stages respectively, the LD amplitudes B1, B2 and B3 are directly output at one time using the LSTM model; For cases with limited input data, a single prediction could compromise accuracy. Therefore, a recursive multi-step prediction approach is adopted initially. However, considering that continuously feeding predicted data into the model would lead to accumulated bias, a direct multi-step prediction approach is used only after a sufficient amount of input data is available. Furthermore, instead of predicting one value per step, a non-arithmic ascending sequence is used to determine the number of predictions per step, further reducing the number of steps and thus minimizing accumulated bias. For cases with a large amount of input data, where the accuracy of the prediction is more assured, a direct multi-step prediction approach is used. This invention addresses both limited and large amounts of input data by devising different prediction methods, thereby improving the accuracy of PD amplitude prediction.
[0013] Preferably, in step S6: for a high-voltage switchgear where a random PD amplitude is currently in the initial stage: if the growth rate between all adjacent PD amplitudes output by the LSTM model is within the range of (0, w1), then the time required for the secondary prediction of the PD amplitude to enter the intermediate stage is H, and a device fault warning is issued: the corresponding switchgear is currently in a weak discharge stage, and it is expected that the discharge will intensify after a period of H. It is recommended to increase the monitoring frequency and check for dirt on the surface of the busbar insulators or poor contact at the connection points before the period of H. Otherwise, the two adjacent PD amplitudes that first appear with a growth rate between them in the range of [w1, w2] are selected in chronological order, and the time when the LSTM model predicts that the first amplitude of the selected two adjacent PD amplitudes will appear is t. ’ The time required for the second-order PD amplitude to enter the intermediate stage is: t ’ -T ’ T ’ The current time is displayed, and a device fault message is issued: The corresponding switchgear is currently in a weak discharge phase, which is expected to last for t seconds. ’ -T ’ Afterwards, an enhanced discharge will occur; it is recommended to wait for a duration of t. ’ -T ’ A power outage inspection was arranged previously; For high-voltage switchgear whose PD amplitude is currently in the middle stage, the estimation method for the time required for the PD amplitude to enter the later stage is the same as the estimation method for the initial stage. The secondary estimation yields that the time required for the PD amplitude of the corresponding high-voltage switchgear to enter the later stage is A, and a equipment fault prompt is issued: The corresponding switchgear is currently in the discharge enhancement stage, and it is expected that a high-intensity discharge will occur after a duration of A. It is recommended to arrange a power outage inspection before the duration of A. For high-voltage switchgear where the PD amplitude is currently in the later stage, the estimation method for the time required for insulation breakdown in the secondary stage is the same as that for the initial stage. The secondary estimation yields a time of C required for insulation breakdown in the corresponding high-voltage switchgear, and issues an equipment fault warning: "High-intensity partial discharge has been detected, posing a high risk of insulation breakdown. Insulation breakdown is expected to occur after time C. It is recommended to immediately take the switchgear out of operation and carry out maintenance."
[0014] An intelligent monitoring system for high-voltage switchgear includes: a status monitoring and judgment module, an input data filtering module, an initial prediction module, a data volume planning module, and a secondary prediction and fault indication module. By using the historical PD amplitude data of the high-voltage switchgear to be predicted by the status monitoring and judgment module set, the current stage of the PD amplitude of different high-voltage switchgear can be determined. The input data filtering module is used to filter out target historical PD amplitude data according to the current stage of the PD amplitude of the high voltage switchgear, and input the filtered target historical PD amplitude data into the LSTM model; The initial estimation module collects the PD amplitude data of a randomly recorded high-voltage switchgear that has completed service, and makes an initial estimation of the time required for the PD amplitude of different high-voltage switchgear to enter the next stage. The required data volume for PD amplitude prediction is planned using the LSTM model through the required data volume planning module. The secondary prediction method is planned through the secondary prediction and fault indication module. Based on the data output by the LSTM model, the time required for the PD amplitude to enter the next stage is estimated, and equipment fault indication is given based on the secondary prediction results.
[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention utilizes an LSTM model to predict PD amplitude to assist in predicting switchgear faults, thereby achieving equipment fault prediction and health management. Considering that the LSTM model itself has difficulty capturing data changes that jump across stages, the current stage of the PD amplitude of different high-voltage switchgear is determined first. Data containing multiple stages is removed from the monitored data before being input into the LSTM model for future PD amplitude prediction. The PD amplitude of each stage exhibits different data patterns, ensuring that the input data is from a single stage, thus reducing the probability of serious deviations in the LSTM model prediction results due to the input of cross-stage data. Based on the initial estimated duration and PD amplitude monitoring interval, this invention plans the amount of data to be predicted for high-voltage switchgear at different stages, i.e., how many PD amplitudes need to be predicted. This increases the likelihood that the model output data will not include cross-stage data, increases the probability of using the LSTM model to predict amplitudes for only a single stage, and further improves the problem that the LSTM model is difficult to capture cross-stage jump data changes and is prone to large deviations in PD amplitude prediction results. The method of using the LSTM model for PD amplitude prediction has been optimized, effectively ensuring the accuracy of PD amplitude prediction results, thereby improving the accuracy of early identification of switchgear risks. Furthermore, for cases with a small amount of data in the input model, the accuracy of the prediction results cannot be guaranteed if the input data is limited and prediction is performed all at once. Therefore, a recursive multi-step prediction method is adopted first. Considering that continuously using recursive multi-step prediction would continuously feed the predicted data into the model as new input data, resulting in accumulated bias in the prediction results, the method of switching to direct multi-step prediction is adopted after the input data is abundant to reduce the accumulated bias. In the recursive multi-step prediction, instead of predicting one value per step, the number of values to be predicted in each step is set by setting a non-arithmetic increasing sequence, which reduces the number of steps and further reduces the accumulated bias. For cases with a large amount of data in the input model, the accuracy of the model prediction results is guaranteed due to the large amount of input data, so direct multi-step prediction is adopted directly. This invention plans different prediction methods for two different cases with small and large amounts of input data, further improving the accuracy of PD amplitude prediction results. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating an intelligent monitoring method for high-voltage switchgear according to the present invention. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] like Figure 1 As shown, an intelligent monitoring method for high-voltage switchgear is provided, including: S1: Collect historical PD amplitude data of the high-voltage switchgear to be predicted, and determine the current stage of the PD amplitude of different high-voltage switchgear. The stage determination method is as follows: obtain the PD amplitudes of the two most recent high-voltage switchgears monitored as Q1 and Q2, set the PD amplitude monitoring interval as t, and calculate the PD amplitude growth rate W of the two most recent high-voltage switchgears monitored according to the formula W=(Q2-Q1) / t. Set the PD amplitude growth rate thresholds for the initial, middle and late stages. The value ranges are (0, w1), [w1, w2], and (w2, +∞), where w1 and w2 are the starting and ending values of the PD amplitude growth rate threshold range in the intermediate stage, respectively. If W is within (0, w1), the PD amplitude of the corresponding high-voltage switchgear is determined to be in the initial stage; if W is within [w1, w2], the PD amplitude of the corresponding high-voltage switchgear is determined to be in the intermediate stage; if W is within (w2, +∞), the PD amplitude of the corresponding high-voltage switchgear is determined to be in the later stage. S2: Based on the current stage of the PD amplitude of the high-voltage switchgear, filter out the target historical PD amplitude data and input the filtered target historical PD amplitude data into the LSTM model: If the current stage of the PD amplitude of the high-voltage switchgear is the initial stage, then the target historical PD amplitude data to be input into the LSTM model is: all historical PD amplitude data of the corresponding high-voltage switchgear that have been monitored; if the current stage of the PD amplitude of the high-voltage switchgear is the middle stage, then the target historical PD amplitude data to be input into the LSTM model is: all the remaining data after removing the PD amplitude data of the initial stage; Example 1: If the PD amplitude of the high-voltage switchgear is currently in the middle stage, and four random PD amplitudes of the corresponding switchgear are obtained at adjacent times, with values of 10pC, 18pC, 50pC and 84pC, and the PD amplitude monitoring interval is 1 hour, the threshold ranges for the PD amplitude growth rate in the initial and middle stages are set as follows: (0, w1) = (0, 10), [w1, w2] = [10, 50]. The calculated growth rates of two adjacent PD amplitudes are 8pC / h, 32pC / h and 34pC / h. 8pC / h is not in the interval [w1, w2], while 32pC / h and 34pC / h are in the interval [w1, w2]. Therefore, the PD amplitudes of 18pC, 50pC and 84pC are the target historical PD amplitude data that need to be input into the LSTM model, and the PD amplitude of 10pC is the PD amplitude data of the initial stage that has been removed. If the PD amplitude of the high-voltage switchgear is currently in the later stage, the target historical PD amplitude data that needs to be input into the LSTM model is: all the remaining data after removing the PD amplitude data of the initial and intermediate stages. S3: Collect PD amplitude data from a randomly selected high-voltage switchgear that has completed its service life. A high-voltage switchgear that has completed its service life refers to switchgear equipment that has reached its designed service life and is no longer in operation. Initially estimate the time required for the PD amplitude of different high-voltage switchgear to enter the next stage. The high-voltage switchgear that has completed its service life and the high-voltage switchgear whose PD amplitude is to be predicted are of the same type. For high-voltage switchgear whose PD amplitude is in the initial stage, the initial estimation method for the time required for the PD amplitude to enter the intermediate stage is as follows: Obtain the PD amplitude growth rate R of the high-voltage switchgear whose PD amplitude is in the initial stage from the two most recently monitored times. Retrieve a random set of PD amplitudes from a high-voltage switchgear that has completed its service life and is in the initial stage. The PD amplitudes in the set are arranged in chronological order of collection time. Calculate the growth rate of every two adjacent PD amplitudes in the set, using the same calculation method as for W. The resulting set of growth rates for every two adjacent PD amplitudes is r = {r1, r2, ..., r...}. m The set contains m+1 PD amplitudes, r m Let r represent the growth rate of the m-th and (m+1)-th PD amplitudes within the set. The growth rate of the m-th and (m+1)-th PD amplitudes is the rate at which the m-th PD amplitude changes to the (m+1)-th PD amplitude. Determine if there exists a value in set r equal to R. If only one value equal to R exists, the value equal to R is r. i Let T1 be the acquisition time of the i-th PD amplitude in the set, and T2 be the time when the PD amplitude of the corresponding high-voltage switchgear that has completed service enters the mid-term stage. Then, the initial estimated duration is H, where H = T2 - T1. If there is more than one value equal to R, the first initial estimated duration is obtained based on the first value in the set that is equal to R. If there is no value equal to R, the initial estimated duration is obtained based on the value in the set with the smallest difference from R. For example: If there exists more than one value equal to R: the value of the first item in the set that is equal to R is r. j If the acquisition time of the j-th PD amplitude in the set is T3, then the initial estimated duration H = T2 - T3 is obtained; If there is no value equal to R: the value in the set with the smallest difference from R is r. e If the acquisition time of the e-th PD amplitude in the set is T4, then the initial estimated duration is H=T2-T4; For high-voltage switchgear with PD amplitude in the mid-stage: The initial estimation method for the duration required for the PD amplitude to enter the late stage is the same as the initial estimation method for the duration required for the PD amplitude of high-voltage switchgear in the initial stage to enter the mid-stage; For high-voltage switchgear with PD amplitude in the late stage: The initial estimation method for the duration required for insulation breakdown is the same as the initial estimation method for the duration required for the PD amplitude of high-voltage switchgear in the initial stage to enter the mid-stage; S4: Plan the way to use the LSTM model for PD amplitude prediction according to the amount of data of the selected target historical PD amplitude data. The methods include Method 1: First perform recursive multi-step prediction and then switch to direct multi-step prediction; Method 2: Perform direct multi-step prediction, specifically including: For a randomly selected high-voltage switchgear with the current stage of PD amplitude being the initial stage among the high-voltage switchgears to be predicted for PD amplitude: A total of k pieces of target historical PD amplitude data that need to be input into the LSTM model are obtained, and the average value of the number of pieces of target historical PD amplitude data that need to be input into the LSTM model for all high-voltage switchgears to be predicted for PD amplitude is L. Compare k and L: If k < L, the method planned to use the LSTM model for PD amplitude prediction for the corresponding switchgear is Method 1; If , the method planned to use the LSTM model for PD amplitude prediction for the corresponding switchgear is Method 2; S5: Plan the amount of data to be predicted for PD amplitude prediction using the LSTM model according to the initially estimated duration: For high-voltage switchgear with PD amplitude in the initial stage: Plan the amount of data to be predicted B1 for PD amplitude prediction using the LSTM model according to the formula B1 = H / t, that is, a total of B1 PD amplitudes need to be predicted, and round B1 down. For high-voltage switchgear with PD amplitude in the mid-stage and late stage, the planning method for the amount of data to be predicted is the same as that in the initial stage, and the amounts of data to be predicted for PD amplitude prediction using the LSTM model for high-voltage switchgear with PD amplitude in the mid-stage and late stage are obtained as B2 and B3 respectively; For a randomly selected high-voltage switchgear with the current stage of PD amplitude being the initial stage: The operation method of Method 1 is: If k < L, set a non-arithmetic increasing sequence with the value of the last term greater than or equal to L: The sequence is {a1, a2,..., a x}, where x represents the number of terms in the sequence, the initial value of the sequence a1 = k, the second value of the sequence a2 = k + 1, a3 = a2 + a1, a x = a x-1 + a x-2The recursive multi-step prediction is performed based on the set sequence: First, one neuron is designed in the output layer of the LSTM model, with each neuron corresponding to a predicted value for one time step. The k historical target PD amplitudes are input into the LSTM model, which outputs the PD amplitudes for the next a2-a1 time steps. These future a2-a1 PD amplitudes are then added to the target historical PD amplitudes to generate the first new target historical PD amplitude. This new target historical PD amplitude is then input into the LSTM model, which outputs the PD amplitudes for the next a3-a2 time steps. These output values are then added to the first new target historical PD amplitude to generate the second new target historical PD amplitude. This second new target historical PD amplitude is then input into the LSTM model, which outputs the PD amplitudes for the next a4-a3 time steps. This process continues, with the output results used as input data to the model until the next a2-a3 time step is output. x -a x-1 The PD amplitude at each time step is calculated, and the output data is added to the input data to obtain the latest input data. Then, the method is switched to direct multi-step prediction: the latest input data is input into the LSTM model, and the LSTM model is used to directly output B1 PD amplitudes at one time. For high-voltage switchgear where the current PD amplitude is in the middle and late stages, the operation method of Method 1 is the same as that in the initial stage. Method 2 operates as follows: For high-voltage switchgear where the PD amplitude is in the initial, middle and late stages respectively, the LSTM model is used to directly output the PD amplitudes B1, B2 and B3 at one time. S6: Based on the data output by the LSTM model, make a secondary estimation of the time required for the PD amplitude to enter the next stage, and provide equipment fault prompts based on the secondary estimation results; For example: For a high-voltage switchgear where a random PD amplitude is currently in the initial stage: If the growth rate between all adjacent PD amplitudes output by the LSTM model is within the interval (0, w1), then the time required for the secondary prediction of the PD amplitude to enter the intermediate stage is H, and a fault warning is issued: The corresponding switchgear is currently in a weak discharge stage, and it is expected that the discharge will intensify after a period of H. It is recommended to increase the monitoring frequency and check for dirt on the surface of the busbar insulators or poor contact at the connection points before the period of H. Otherwise, the two adjacent PD amplitudes with the first occurrence of a growth rate between them in the [w1, w2] are selected in chronological order. The time when the LSTM model predicts that the first of the selected two adjacent PD amplitudes will appear is t. ’ The time required for the second-order PD amplitude to enter the intermediate stage is: t ’ -T ’ T ’The current time is displayed, and a device fault message is issued: The corresponding switchgear is currently in a weak discharge phase, which is expected to last for t seconds. ’ -T ’ Afterwards, an enhanced discharge will occur; it is recommended to wait for a duration of t. ’ -T ’ A power outage inspection was scheduled previously.
[0019] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
Claims
1. An intelligent monitoring method for high-voltage switchgear, characterized in that: Includes the following steps: S1: Collect historical PD amplitude data of the high-voltage switchgear that has been monitored and is to be predicted for PD amplitude, and determine the current stage of PD amplitude of different high-voltage switchgear. S2: Based on the current stage of the PD amplitude of the high-voltage switchgear, select the target historical PD amplitude data and input the selected target historical PD amplitude data into the LSTM model; S3: Collect the PD amplitude data of a randomly recorded high-voltage switchgear that has completed service, and make an initial estimate of the time required for the PD amplitude of different high-voltage switchgear to enter the next stage. S4: Based on the amount of historical PD amplitude data selected from the target data, plan the method of using the LSTM model to predict the PD amplitude. The method includes: Method 1: first perform recursive multi-step prediction, and then switch to direct multi-step prediction; Method 2: perform direct multi-step prediction. S5: Based on the initial estimated duration, plan the amount of data required for PD amplitude prediction using the LSTM model; S6: Based on the data output by the LSTM model, make a secondary prediction of the time required for the PD amplitude to enter the next stage, and provide equipment fault prompts based on the secondary prediction results.
2. The intelligent monitoring method for high-voltage switchgear according to claim 1, characterized in that: In step S1: The current stage of the PD amplitude of the different high-voltage switchgear includes the initial stage, the middle stage and the late stage. The stage is determined by: obtaining the PD amplitude of the high-voltage switchgear in the two most recent monitoring and the set monitoring interval time, calculating the growth rate of the PD amplitude of the high-voltage switchgear in the two most recent monitoring, setting the threshold range of the PD amplitude growth rate for the three stages respectively, and determining the current stage of the PD amplitude of the different high-voltage switchgear based on the comparison between the calculation results and the threshold ranges.
3. The intelligent monitoring method for high-voltage switchgear according to claim 2, characterized in that: In step S2: If the current stage of the PD amplitude of the high-voltage switchgear is the initial stage, the target historical PD amplitude data that needs to be input into the LSTM model is: all historical PD amplitude data of the corresponding high-voltage switchgear that have been monitored. If the PD amplitude of the high-voltage switchgear is currently in the middle stage, the target historical PD amplitude data to be input into the LSTM model is: all the remaining data after removing the PD amplitude data of the initial stage; if the PD amplitude of the high-voltage switchgear is currently in the later stage, the target historical PD amplitude data to be input into the LSTM model is: all the remaining data after removing the PD amplitude data of the initial and middle stages.
4. The intelligent monitoring method for high-voltage switchgear according to claim 3, characterized in that: In step S3: For high-voltage switchgear with PD amplitude in the initial stage, the initial estimation method for the time required for PD amplitude to enter the intermediate stage is as follows: Obtain the PD amplitude growth rate R of the high-voltage switchgear with PD amplitude in the initial stage from the two most recent monitoring data. Retrieve a random set of PD amplitudes of high-voltage switchgear in the initial stage that has completed service. The PD amplitudes in the set are arranged in order from the time of acquisition to the time of acquisition. Calculate the growth rate of every two adjacent PD amplitudes in the set. The calculation method is the same as the calculation method of W. The set of growth rates of every two adjacent PD amplitudes in the set is obtained as r = {r1, r2, ..., r...} m The set contains m+1 PD amplitudes, r m Let R represent the growth rate of the m-th and (m+1)-th PD values within the set. Determine if there exists a value in set r equal to R. If only one value equal to R exists, then the value equal to R is r. i Let T1 be the acquisition time of the i-th PD amplitude in the set, and T2 be the time when the PD amplitude of the corresponding high-voltage switchgear that has completed service enters the mid-term stage. Then, the initial estimated duration is H, where H = T2 - T1. If there is more than one value equal to R, the first initial estimated duration is obtained based on the first value in the set that is equal to R. If there is no value equal to R, the initial estimated duration is obtained based on the value in the set with the smallest difference from R.
5. The intelligent monitoring method for high-voltage switchgear according to claim 4, characterized in that: For high-voltage switchgear with PD amplitude in the middle stage: the initial estimation method for the time required for PD amplitude to enter the later stage is the same as the initial estimation method for the time required for PD amplitude to enter the middle stage for high-voltage switchgear with PD amplitude in the early stage; for high-voltage switchgear with PD amplitude in the later stage: the initial estimation method for the time required for insulation breakdown is the same as the initial estimation method for the time required for PD amplitude to enter the middle stage for high-voltage switchgear with PD amplitude in the early stage.
6. The intelligent monitoring method for high-voltage switchgear according to claim 5, characterized in that: In step S4: For a high-voltage switchgear for which PD amplitude prediction is to be performed, randomly select a high-voltage switchgear in the initial stage of the current PD amplitude phase: A total of k target historical PD amplitude data to be input into the LSTM model are obtained. The average number of target historical PD amplitude data to be input into the LSTM model for all high-voltage switchgears for which PD amplitude prediction is to be performed is L. Compare k and L: If k < L, the method for PD amplitude prediction using the LSTM model planned for the corresponding switchgear is Method 1; if , the method for PD amplitude prediction using the LSTM model planned for the corresponding switchgear is Method 2.
7. The intelligent monitoring method for high-voltage switchgear according to claim 6, characterized in that: In step S5: For high-voltage switchgear with PD amplitude in the initial stage: According to the formula B1=H / t, the amount of data to be predicted using the LSTM model for PD amplitude prediction is planned, B1, and B1 is rounded down; For high-voltage switchgear with PD amplitude in the middle and late stages, the planning method for the amount of data to be predicted is the same as that for the initial stage, resulting in the amount of data to be predicted using the LSTM model for PD amplitude prediction for high-voltage switchgear with PD amplitude in the middle and late stages being B2 and B3, respectively.
8. The intelligent monitoring method for high-voltage switchgear according to claim 7, characterized in that: For a high-voltage switchgear where the current stage of a random PD amplitude is the initial stage: The operation mode of the first method is as follows: If k < L, set a non-arithmetic increasing sequence with the value of the last term greater than or equal to L: The sequence is {a1, a2,..., a x}, where x represents the number of sequence terms. The initial value of the sequence a1 = k, the second term value of the sequence a2 = k + 1, a3 = a2 + a1, a x = a x-1 + a x-2 . Complete recursive multi-step prediction according to the set sequence: First, design 1 neuron in the output layer of the LSTM model. Each neuron corresponds to the predicted value of a time step. Input k target historical PD amplitudes into the LSTM model. The LSTM model outputs the PD amplitudes of the future a2 - a1 time steps. Then add the output PD amplitudes to the target historical PD amplitudes to generate the first new target historical PD amplitudes. Input the new target historical PD amplitudes into the LSTM model. The LSTM model outputs the PD amplitudes of the future a3 - a2 time steps. Then add the output value to the first new target historical PD amplitudes to generate the second new target historical PD amplitudes. Input the second new target historical PD amplitudes into the LSTM model and output the PD amplitudes of the future a4 - a3 time steps. And so on, use the output result as the input data and input it into the model until the PD amplitudes of the future a x - a x-1 time steps are output. Add the output data to the input data to obtain the latest input data. Then switch to the direct multi-step prediction mode: Input the latest input data into the LSTM model, and the LSTM model directly outputs B1 PD amplitudes at one time. For high-voltage switchgears where the current stage of the PD amplitude is the middle and late stages, the operation mode of the first method is the same as that in the initial stage; The operation method of the second method is as follows: for high-voltage switchgear with PD amplitude in the initial, middle and late stages respectively, the LSTM model is used to directly output the PD amplitudes B1, B2 and B3 respectively at one time.
9. The intelligent monitoring method for high-voltage switchgear according to claim 8, characterized in that: In step S6: For a high-voltage switchgear where a random PD amplitude is currently in the initial stage: If the growth rate between all adjacent PD amplitudes output by the LSTM model is within the interval (0, w1), then the time required for the secondary prediction of the PD amplitude to enter the intermediate stage is H, and a device fault warning is issued: The corresponding switchgear is currently in a weak discharge stage, and it is expected that the discharge will intensify after a period of H. It is recommended to increase the monitoring frequency and check for dirt on the surface of the busbar insulators or poor contact at the connection points before the period of H. Otherwise, the two adjacent PD amplitudes that first appear with a growth rate between them in the [w1, w2] are selected in chronological order. The time when the LSTM model predicts that the first amplitude of the selected two adjacent PD amplitudes will appear is t. ’ The time required for the second-order PD amplitude to enter the intermediate stage is: t ’ -T ’ T ’ Set the current time and issue a device malfunction warning; For high-voltage switchgear whose PD amplitude is currently in the middle stage, the estimation method for the time required for the PD amplitude to enter the later stage is the same as the estimation method for the initial stage. The secondary estimation yields that the time required for the PD amplitude of the corresponding high-voltage switchgear to enter the later stage is A, and a equipment fault prompt is issued: The corresponding switchgear is currently in the discharge enhancement stage, and it is expected that a high-intensity discharge will occur after a duration of A. It is recommended to arrange a power outage inspection before the duration of A. For high-voltage switchgear where the PD amplitude is currently in the later stage, the estimation method for the time required for insulation breakdown in the secondary stage is the same as that for the initial stage. The secondary estimation yields a time of C required for insulation breakdown in the corresponding high-voltage switchgear, and issues an equipment fault warning: "High-intensity partial discharge has been detected, posing a high risk of insulation breakdown. Insulation breakdown is expected to occur after time C. It is recommended to immediately take the switchgear out of operation and carry out maintenance." 10. An intelligent monitoring system for high-voltage switchgear, comprising the intelligent monitoring method for high-voltage switchgear as described in any one of claims 1-9, characterized in that: The system includes: a status monitoring and judgment module, an input data filtering module, an initial estimation module, a data volume planning module, and a secondary estimation and fault indication module. By using the historical PD amplitude data of the high-voltage switchgear to be predicted by the status monitoring and judgment module set, the current stage of the PD amplitude of different high-voltage switchgear can be determined. The input data filtering module is used to filter out target historical PD amplitude data according to the current stage of the PD amplitude of the high voltage switchgear, and input the filtered target historical PD amplitude data into the LSTM model; The initial estimation module collects the PD amplitude data of a randomly recorded high-voltage switchgear that has completed service, and makes an initial estimation of the time required for the PD amplitude of different high-voltage switchgear to enter the next stage. The required data volume for PD amplitude prediction is planned using the LSTM model through the required data volume planning module. The secondary prediction method is planned through the secondary prediction and fault indication module. Based on the data output by the LSTM model, the time required for the PD amplitude to enter the next stage is estimated, and equipment fault indication is given based on the secondary prediction results.