Abnormal current warning method for distribution cabinet based on big data

Abstract

The present invention relates to the field of power monitoring technology, and in particular to a method for early warning of current anomalies in distribution cabinets based on big data. When a predicted fault type is not predicted, a first preset value is used as a sampling time interval, and current data is collected at a lower frequency, thereby saving equipment power consumption and reducing the cost of data collection and transmission. When it is difficult to accurately predict the fault type based on the first prediction sequence, the sampling time interval is adjusted, and current data is collected at a higher frequency in a second time period. If the accuracy of the fault type prediction is not improved, the collection of operating status information of the target distribution cabinet is started in a third time period, and then fault prediction is performed. Data collection and fault prediction are performed in a staged manner. Compared with the method of continuously collecting data at a high frequency in the prior art, the cost of abnormal early warning can be greatly reduced while ensuring the accuracy of fault prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of power monitoring, and in particular to a method for early warning of abnormal current in a distribution cabinet based on big data. Background Art

[0002] As a key component of the power system, distribution cabinets (DPCs) are responsible for distributing and controlling electrical energy. During power system operation, the current status in the DPCs directly reflects the health of the power equipment. Real-time monitoring of DPC current and prompt detection of abnormalities are crucial for ensuring the safe and stable operation of the power system. For example, current overloads can cause equipment overheating, damage, and even fires. Abnormal current fluctuations can indicate problems such as poor line contact and electrical equipment failure.

[0003] With the development of science and technology, intelligent monitoring technology is gradually being applied to the field of current monitoring in power distribution cabinets. For example, current transformers, sensors, and other devices are used to collect current signals in real time. The signals are then transmitted to the monitoring center server through a data acquisition system for analysis and prediction, thus realizing early warning functions.

[0004] However, the above method usually collects current signals at a relatively high fixed frequency, and the transmission pressure of the current signal is relatively large. In the scenario where the server of the monitoring center supports current monitoring of multiple distribution cabinets at the same time, there may even be congestion in the data transmission line of the server, resulting in the inability to obtain the current signal in time or the loss of the current signal, which in turn leads to a decrease in the accuracy of the current monitoring of the distribution cabinet.

[0005] In addition, relying solely on current signals for analysis and prediction may lead to misjudgment of faults. Therefore, in the existing method, the operating status information of the distribution cabinet is also introduced to improve the accuracy of fault prediction through the combination of multi-dimensional data. Obviously, the introduction of operating status information will further increase the burden of data transmission, resulting in a higher cost for abnormal current warning of the distribution cabinet.

[0006] Therefore, how to reduce the cost of abnormal current warning in distribution cabinets has become an urgent problem to be solved. Summary of the Invention

[0007] In response to the above technical problems, the technical solution adopted by the present invention is a distribution cabinet current abnormality early warning method based on big data, which includes the following steps:

[0008] S1, based on the first acquisition sequence formed by the current acquisition values ​​corresponding to each first preset time point of the target distribution cabinet in the first time period, predict and obtain the first prediction sequence formed by the first current prediction values ​​corresponding to each second preset time point of the target distribution cabinet in the fourth time period, wherein the time interval between adjacent first preset time points is the first preset value, the time interval between adjacent second preset time points is the second preset value, and the first preset value is an integer multiple of the second preset value.

[0009] S2: If the first prediction sequence meets the first preset condition, then according to the time interval corresponding to the second preset value, sample to obtain a second acquisition sequence formed by the current acquisition values ​​of the target distribution cabinet corresponding to each second preset time point in the second time period.

[0010] S3, sampling the second acquisition sequence according to the time interval corresponding to the first preset value, to obtain a third acquisition sequence formed by the current acquisition values ​​of the target distribution cabinet corresponding to each first preset time point in the second time period.

[0011] S4. If the second acquisition sequence and the third acquisition sequence meet the second preset condition, a fourth acquisition matrix formed by the current acquisition values ​​and operating status information corresponding to each second preset time point of the target distribution cabinet in the third time period is obtained.

[0012] S5. Predict, based on the fourth acquisition matrix, fourth predicted probabilities of the target distribution cabinet corresponding to the M fault types.

[0013] S6. Obtain abnormal warning information corresponding to the target distribution cabinet according to the M fourth prediction probabilities.

[0014] The present invention has at least the following beneficial effects: when the predicted fault type is not predicted, the first preset value is used as the sampling time interval between adjacent preset time points, and current data is collected at a lower frequency, which saves equipment power consumption and reduces the cost of data collection and transmission. When it is difficult to accurately predict the fault type based on the first prediction sequence, the sampling time interval is adjusted, and current data is collected at a higher frequency in the second time period. If the accuracy of the fault type prediction is not improved after the sampling time interval is adjusted, the collection of the operating status information of the target distribution cabinet is started in the third time period, and then subsequent fault prediction is performed, thereby performing data collection and fault prediction in a staged manner. Compared with the method of continuously collecting data at a high frequency in the prior art, the cost of abnormal warning can be greatly reduced while ensuring the accuracy of fault prediction. Especially in the scenario where the data transmission bandwidth of the server is limited, the server can support current abnormality monitoring and early warning of more distribution cabinets. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 A flowchart of a method for early warning of abnormal current in a distribution cabinet based on big data is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0018] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It is understood that, where appropriate, the above-mentioned terms used to distinguish similar objects can be interchanged so that the present invention can also implement other embodiments other than the above-mentioned illustrated embodiments or described embodiments. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0019] This embodiment provides a method for early warning of abnormal current in a distribution cabinet based on big data, such as Figure 1 FIG. 1 is a flow chart of a method for early warning of abnormal current in a distribution cabinet based on big data provided by an embodiment of the present invention. The method for early warning of abnormal current in a distribution cabinet based on big data includes the following steps:

[0020] S1, based on the first acquisition sequence formed by the current acquisition values ​​corresponding to each first preset time point of the target distribution cabinet in the first time period, predict and obtain the first prediction sequence formed by the first current prediction values ​​corresponding to each second preset time point of the target distribution cabinet in the fourth time period, wherein the time interval between adjacent first preset time points is the first preset value, the time interval between adjacent second preset time points is the second preset value, and the first preset value is an integer multiple of the second preset value.

[0021] The target distribution cabinet is the one that requires current anomaly monitoring and early warning. Data acquisition equipment is installed in the target distribution cabinet to collect current values ​​and operating status information based on actual conditions. This data acquisition equipment includes various devices such as ammeters, temperature sensors, humidity sensors, and vibration sensors.

[0022] The first time period includes N1 first preset time points. Accordingly, the first acquisition sequence can be considered a vector of size 1×N1. The first preset value can represent the time interval when data is collected in the first time period, and N1 can represent the amount of data collected in the first time period. The fourth time period includes N2 second preset time points. Accordingly, the first prediction sequence can be considered a vector of size 1×N2. The second preset value can represent the time interval when data is predicted in the fourth time period, and N2 can represent the amount of data predicted in the fourth time period.

[0023] N1 and N2 are both integers greater than 0. The first preset value is an integer multiple of the second preset value. Taking the integer multiple J as an example, N2 = N1 × J, where J is an integer greater than 0. The specific values ​​of N1, N2, and J can be set by the implementer based on actual conditions. In this embodiment, J can be set to 2.

[0024] The time lengths corresponding to the first time period, the second time period, the third time period, and the fourth time period are all the same. The end time of the first time period is the same as the start time of the second time period, the end time of the second time period is the same as the start time of the third time period, and the end time of the third time period is the same as the start time of the fourth time period.

[0025] Based on the data actually collected in the first time period, the relevant data of the target distribution cabinet in the subsequent second time period, third time period and fourth time period are predicted as the data basis for early warning of current anomalies.

[0026] In one embodiment, S1 includes the following steps:

[0027] S11 , starting from the start time point of the first time period, collecting current collection values ​​of the target distribution cabinet corresponding to each first preset time point in the first time period.

[0028] S12: splicing the current collection values ​​corresponding to the first preset time points of the target distribution cabinet in the first time period in chronological order to obtain a first collection sequence.

[0029] S13: Input the first acquisition sequence into the trained time series prediction model to obtain the first current prediction value of the target distribution cabinet corresponding to each second preset time point in the fourth time period.

[0030] S14: splicing the first current prediction values ​​of the target distribution cabinet corresponding to each second preset time point in the fourth time period in chronological order to obtain a first prediction sequence.

[0031] Among them, the time series prediction model can adopt a recurrent neural network model, a long short-term memory network model, a time domain convolutional network model, etc.

[0032] Specifically, this embodiment takes the time domain convolutional network model adopted in the time series prediction model as an example. In order to be compatible with the prediction of subsequent high-frequency current acquisition values ​​and the prediction of subsequent operating status information, this embodiment sets the input size of the time series prediction model to (I+1)×(J×N1), where I is the number of operating status types. The operating status types may include temperature, humidity in the cabinet, target size, etc. The implementer can set it according to actual conditions. J is the ratio of the first preset value to the second preset value, which is also the ratio of N2 to N1.

[0033] In order to ensure the size consistency of the input data of the time series prediction model, the first acquisition sequence is expanded from a vector of size 1×N1 to a matrix of size (I+1)×(J×N1). First, J-1 mask values ​​are inserted between two adjacent current acquisition values ​​in the first acquisition sequence, and J-1 mask values ​​are inserted after the last current acquisition value in the first acquisition sequence to form a vector of size 1×(J×N1). For any mask value, the mask value is assigned according to the current acquisition value corresponding to the first preset time point closest to the mask value in the first acquisition sequence. Since the operation status information is not collected at the current time, the matrix corresponding to the operation status information is set to a preset zero vector matrix of size I×(J×N1). The vector of size 1×(J×N1) is then concatenated with the zero vector matrix of size I×(J×N1) to obtain a matrix of size (I+1)×(J×N1).

[0034] The output size of the time series prediction model is set to a vector of size (I+1)×1, which can represent the predicted current value and predicted state information corresponding to the target distribution cabinet at a single second preset time point.

[0035] The 1×(J×N1)-sized vector corresponds to the first row of the input (I+1)×(J×N1)-sized matrix. Correspondingly, the values ​​in the first row of the output (I+1)×1-sized vector represent the corresponding predicted current values, and the values ​​in rows 2 through (I+1) represent the corresponding predicted state information.

[0036] The time series prediction model uses an iterative prediction method to make predictions. For example, the matrix of size (I+1)×(J×N1) corresponding to the first acquisition sequence is used as input, and the prediction result corresponding to the first second preset time point in the second time period is output. Then, the vectors from the second column to the J×N1 column in the matrix of size (I+1)×(J×N1) corresponding to the first acquisition sequence are concatenated with the prediction result corresponding to the first second preset time point in the second time period to obtain a new input matrix, and the prediction result corresponding to the second preset time point in the second time period is output. Then, the vectors from the second column to the J×N1 column in the matrix of size (I+1)×(J×N1) corresponding to the first acquisition sequence are concatenated with the prediction result corresponding to the first second preset time point in the second time period to obtain a new input matrix. The vectors from the 3rd column to the J×N1 column in the matrix of size (I+1)×(J×N1) are concatenated with the prediction results corresponding to the 1st and 2nd second time points in the second time period to obtain a new input matrix, and the prediction result corresponding to the 3rd in the second time period is output, and so on, until the first current prediction values ​​corresponding to the N2 second preset time points in the second time period are obtained, and so on, the first current prediction values ​​corresponding to the N2 second preset time points in the third time period and the first current prediction values ​​corresponding to the N2 second preset time points in the fourth time period are obtained.

[0037] It can be known that in order to ensure the generalization of the time series prediction model, the implementer should, according to the above-mentioned prediction reasoning process, use the matrix obtained by splicing the vector expanded from the current acquisition value obtained by sampling at the time interval of the first preset value and the zero vector matrix as the first training sample when constructing the training set of the time series prediction model, and use the matrix obtained by splicing the current acquisition value obtained by sampling at the time interval of the second preset value and the operating status information as the second training sample. The training loss of the time series prediction model can adopt the mean square error loss function.

[0038] In the above, the actual current in the target distribution cabinet is sampled with the first preset value as the sampling time interval to obtain a first acquisition sequence. The current data of the target distribution cabinet in the subsequent second time period, third time period and fourth time period are predicted based on the first acquisition data to obtain the first prediction sequence corresponding to the fourth time period, which serves as the data basis for setting the subsequent data collection method to provide early warning of current abnormalities.

[0039] S2: If the first prediction sequence meets the first preset condition, then according to the time interval corresponding to the second preset value, sample to obtain a second acquisition sequence formed by the current acquisition values ​​of the target distribution cabinet corresponding to each second preset time point in the second time period.

[0040] Based on the first prediction sequence for the target distribution cabinet in the fourth time period, the probability of the target distribution cabinet belonging to each fault type is predicted and analyzed. The reliability of the fault prediction results is analyzed based on the predicted probabilities. If the reliability is high, the fault prediction results corresponding to the first prediction sequence can be directly used to provide early warning for the target distribution cabinet. If the reliability is low and the fault type cannot be accurately predicted based on the first prediction sequence, the sampling interval is adjusted, and current data is collected at a higher frequency in the second time period to obtain richer data as the data basis for early warning of current anomalies.

[0041] In one embodiment, S2 includes the following steps:

[0042] S21, splicing the first prediction sequence and the preset zero vector matrix to obtain a first splicing result.

[0043] S22: Input the first splicing result into a trained fault prediction model to predict first prediction probabilities corresponding to the target distribution cabinet for M fault types.

[0044] S23: If the maximum value of the M first prediction probabilities is less than a preset probability threshold, it is determined that the first prediction sequence meets a first preset condition.

[0045] Where M represents the number of fault types and is a positive integer. Fault types may include poor contact of a switching element, current transformer fault, load device fault, cable damage fault, etc. Implementers may set these based on actual conditions. It should be noted that in this embodiment, the fault type also includes no fault.

[0046] The fault prediction model can use a classification network model, which can include convolutional layers and fully connected layers. The convolutional layer is used to extract features from the input data. The extracted features are flattened and then input into the fully connected layer for mapping. The mapping values ​​corresponding to M fault types are obtained. The M mapping values ​​are normalized using the softmax function to obtain the predicted probabilities corresponding to the M fault types.

[0047] Specifically, the input size of the fault prediction model is (I+1)×N2. In this step, the first prediction sequence is a vector of size 1×N2, and the zero vector matrix is ​​a zero matrix of size I×N2. The first prediction sequence is taken as the first row and is spliced ​​with the zero vector matrix row by row to obtain a first splicing result of size (I+1)×N2.

[0048] It can be known that in order to ensure the generalization of the fault prediction model, the implementer should, when constructing the training set of the time series prediction model, use the matrix obtained by concatenating the current sampling values ​​obtained by sampling at time intervals of the second preset value and the zero vector matrix as the third training sample, and use the matrix obtained by concatenating the current sampling values ​​obtained by sampling at time intervals of the second preset value and the sample state information as the fourth training sample. The training loss of the fault prediction model adopts the cross entropy loss function, and the label is set to the one-hot encoding of the fault type corresponding to the input matrix.

[0049] The first preset condition is used to determine whether the current value prediction and the fault type prediction based on the current collection value collected at the time interval using the first preset value are reliable.

[0050] Specifically, if the maximum value among the M first predicted probabilities is greater than or equal to a preset probability threshold, it can be considered that the fault type prediction result is highly likely to belong to a certain fault type. In this case, the fault type prediction result is considered reliable. The fault type corresponding to the maximum value among the M first predicted probabilities is then used as the third predicted fault type of the target power distribution cabinet, and abnormality warning information is generated based on the third predicted fault type. It should be noted that if the third predicted fault type is no fault, abnormality warning information is not generated.

[0051] If the maximum value among the M first prediction probabilities is less than the preset probability threshold, it can be considered that the probability of the fault type prediction result belonging to any fault type is not high. At this time, the fault type prediction result is considered unreliable, and it is necessary to adjust the sampling time interval and collect current data at a higher frequency in the second time period to obtain richer data as the data basis for early warning of current anomalies.

[0052] The preset probability threshold may be set to 0.8, and implementers may adjust the preset probability threshold according to actual conditions.

[0053] In a specific embodiment, S2 further includes the following steps:

[0054] S24 , starting from the start time point of the second time period, collecting current collection values ​​of the target distribution cabinet corresponding to each second preset time point in the second time period.

[0055] S25 , splicing the current collection values ​​corresponding to the second preset time points of the target distribution cabinet in the second time period in chronological order to obtain a second collection sequence.

[0056] The second acquisition sequence includes N2 current acquisition values, and the second acquisition sequence can be regarded as a vector of size 1×N2.

[0057] As described above, based on the first prediction sequence of the target distribution cabinet in the fourth time period, the probability of the target distribution cabinet belonging to each fault type is predicted and analyzed, and the reliability of the fault prediction result is analyzed based on the predicted probability. If the reliability is low and it is difficult to accurately predict the fault type based on the first prediction sequence, current data collection is performed at a higher frequency in the second time period, so as to obtain richer data as the data basis for early warning of current anomalies.

[0058] S3, sampling the second acquisition sequence according to the time interval corresponding to the first preset value, to obtain a third acquisition sequence formed by the current acquisition values ​​of the target distribution cabinet corresponding to each first preset time point in the second time period.

[0059] The second acquisition sequence is sampled according to the time interval corresponding to the first preset value to obtain the third acquisition sequence, that is, the k×J+1th current acquisition value in the second acquisition sequence is selected and added to the third acquisition sequence, where k=0, 1, 2, ...

[0060] The difference between the third acquisition sequence and the second acquisition sequence lies in the different sampling frequencies and data volumes. By comparing and analyzing the second prediction sequence and the third prediction sequence, we can measure the change in the predicted probability of the fault type after the sampling frequency increases, so as to determine the specific practices for subsequent data acquisition and fault prediction.

[0061] S4. If the second acquisition sequence and the third acquisition sequence meet the second preset condition, a fourth acquisition matrix formed by the current acquisition values ​​and operating status information corresponding to each second preset time point of the target distribution cabinet in the third time period is obtained.

[0062] In a specific embodiment, S4 includes the following steps:

[0063] S41: Input the second acquisition sequence into the trained time series prediction model to obtain the second current prediction value of the target distribution cabinet corresponding to each second preset time point in the fourth time period.

[0064] S42: splicing the second current prediction values ​​of the target distribution cabinet corresponding to each second preset time point in the fourth time period in chronological order to obtain a second prediction sequence.

[0065] S43: Obtain second prediction probabilities corresponding to the M fault types according to the second prediction sequence.

[0066] S44: Input the third acquisition sequence into the trained time series prediction model to obtain third current prediction values ​​of the target distribution cabinet corresponding to each second preset time point in the fourth time period.

[0067] S45 , splicing the third current prediction values ​​of the target distribution cabinet corresponding to each second preset time point in the fourth time period in chronological order to obtain a third prediction sequence.

[0068] S46 , obtaining third prediction probabilities corresponding to the M fault types respectively according to the third prediction sequence.

[0069] S47 , judging whether the second acquisition sequence and the third acquisition sequence meet a second preset condition according to the second prediction probabilities and the third prediction probabilities respectively corresponding to the M fault types.

[0070] The second acquisition sequence is a vector of a size of 1×N2, and the second prediction sequence is a vector of a size of 1×N2.

[0071] The third acquisition sequence is a vector of a size of 1×N1, and the third prediction sequence is a vector of a size of 1×N2.

[0072] Specifically, the reasoning process of the trained time series prediction model can be found in the corresponding part above of this embodiment and will not be repeated here.

[0073] According to the concatenation of the second acquisition sequence and the zero vector matrix of size I×(J×N1), a matrix of size (I+1)×(J×N1) corresponding to the second acquisition sequence is obtained. Iterative prediction is performed through the trained timing prediction model, and the second current prediction values ​​corresponding to the N2 second preset time points in the third time period and the second current prediction values ​​corresponding to the N2 second preset time points in the fourth time period are output respectively, thereby obtaining the second prediction sequence corresponding to the fourth time period.

[0074] Correspondingly, referring to the prediction process according to the first acquisition sequence, the third current prediction values ​​corresponding to the N2 second preset time points in the four time periods are predicted according to the third acquisition sequence, thereby obtaining the third prediction sequence corresponding to the fourth time period.

[0075] Furthermore, by comparing and analyzing the second and third prediction sequences, the change in the predicted probability of the fault type after increasing the sampling frequency is measured, and a determination is made as to whether the second and third acquisition sequences meet the second preset condition. The second preset condition is used to determine whether increasing the current sampling frequency significantly improves the reliability of fault type prediction, and furthermore, whether more complete data dimensions are needed for fault type prediction, thereby determining the specific approach for subsequent data collection and fault prediction.

[0076] In a specific embodiment, S43 includes the following steps:

[0077] S431 , concatenate the second prediction sequence and the preset zero vector matrix to obtain a second concatenation result.

[0078] S432: Input the second splicing result into the trained fault prediction model to predict the second prediction probabilities of the target distribution cabinet corresponding to the M fault types.

[0079] In one specific embodiment, S46 includes the following steps:

[0080] S461: Splice the third prediction sequence and the preset zero vector matrix to obtain a third splicing result.

[0081] S462: Input the third concatenation result into the trained fault prediction model to predict third prediction probabilities corresponding to the target distribution cabinet for the M fault types.

[0082] The reasoning process of the trained fault prediction model is described in the corresponding part above in this embodiment and will not be repeated here.

[0083] In a specific embodiment, S47 includes the following steps:

[0084] S471: Determine the fault type corresponding to the maximum second predicted probability among the M fault types as the first predicted fault type corresponding to the target power distribution cabinet.

[0085] S472: Determine the fault type corresponding to the largest third predicted probability among the M fault types as the second predicted fault type corresponding to the target power distribution cabinet.

[0086] S473: Determine the difference between the maximum third predicted probability and the maximum second predicted probability as the predicted probability difference.

[0087] S474: If the first predicted fault type and the second predicted fault type are inconsistent, or the predicted probability difference is less than a preset probability difference threshold, determine that the second acquisition sequence and the third acquisition sequence meet a second preset condition.

[0088] Among them, if the first predicted fault type and the second predicted fault type are inconsistent, it means that after the sampling frequency increases, the prediction result of the fault type changes, but the change in the prediction result is too large, then the reliability of the change in the prediction result is low, and a more complete data dimension is required to perform a secondary prediction of the fault type.

[0089] If the predicted probability difference is less than the preset probability difference threshold, it means that after the sampling frequency increases, the change in the predicted result of the fault type is too small, and the reliability of the change in the prediction result is low. A more complete data dimension is needed to perform in-depth prediction of the fault type.

[0090] If the first predicted fault type is consistent with the second predicted fault type, and the predicted probability difference is greater than or equal to the preset probability difference threshold, it means that after the sampling frequency increases, the reliability of the change in the prediction result of the fault type is higher. The second predicted fault type can be used as the fault prediction result of the target distribution cabinet to further generate abnormal warning information.

[0091] In a specific embodiment, S4 further includes the following steps:

[0092] S401, starting from the starting time point of the third time period, collect the current collection values ​​corresponding to each second preset time point in the third time period of the target distribution cabinet, and the corresponding operating status information, wherein each operating status information includes status data corresponding to I operating status types, and I is a positive integer.

[0093] S402 : splicing the current collection values ​​corresponding to the second preset time points of the target distribution cabinet in the third time period in chronological order to obtain a fourth collection sequence.

[0094] S403 , performing vectorization processing on each piece of running status data to obtain a state vector corresponding to each piece of running status data.

[0095] S404 , determining the position of each state vector in the state matrix according to a time sequence and a preset sequence of operating state types.

[0096] S405 , obtaining a state matrix according to the position of each state vector in the state matrix.

[0097] S406 , concatenate the fourth acquisition sequence and the state matrix to obtain a fourth acquisition matrix.

[0098] In addition to collecting current values, the data acquisition device also collects operating status information of the target distribution cabinet in the third time period, providing multi-dimensional data support for determining the fault type of the target distribution cabinet.

[0099] Each operating state data is converted into a corresponding state vector according to a vector conversion method to unify the data scale and facilitate the fusion of data of different dimensions. In this embodiment, a suitable vector conversion method is selected according to the type of operating state data. Those skilled in the art know that any vector conversion method in the prior art falls within the scope of protection of the present invention, for example, the Word2Vec algorithm, the One-Hot algorithm, the convolutional neural network, etc., which will not be described in detail here.

[0100] In the state matrix, each operating state type is used as a row key of the state matrix, and each second time point in the third time period is used as a column key of the state matrix. The position of each state vector in the state matrix is ​​determined according to the chronological order and the preset order of the operating state types, and the value of each state vector is filled in the corresponding position to obtain the state matrix. The state matrix can simultaneously represent the current condition and operating state of the target distribution cabinet in the third time period.

[0101] The fourth acquisition sequence with a size of 1×N2 is used as the first row of the fourth acquisition matrix and is spliced ​​with the state matrix with a size of I×N2 to obtain a fourth acquisition matrix with a size of (I+1)×N2 to ensure the size uniformity of the model input.

[0102] As described above, by comparing and analyzing the second prediction sequence and the third prediction sequence, the change in the predicted probability of the fault type after the sampling frequency is increased is measured, and it is determined whether increasing the sampling frequency of the current value has a significant improvement in the reliability of the fault type prediction. If there is no significant improvement, the current data and operating status information are collected simultaneously in the third time period to predict the fault type based on a more complete data dimension, thereby ensuring that the fault prediction result has a high accuracy.

[0103] S5. Predict, based on the fourth acquisition matrix, fourth predicted probabilities of the target distribution cabinet corresponding to the M fault types.

[0104] The fourth acquisition matrix with a size of (I+1)×N2 is input into the trained fault prediction model to predict the fourth prediction probabilities of the target distribution cabinet corresponding to the M fault types.

[0105] S6. Obtain abnormal warning information corresponding to the target distribution cabinet according to the M fourth prediction probabilities.

[0106] In a specific embodiment, S6 includes the following steps:

[0107] S61: Determine the fault type corresponding to the largest fourth predicted probability as the third predicted fault type corresponding to the target power distribution cabinet.

[0108] S62: Generate abnormal warning information according to the third predicted fault type.

[0109] The information template is populated with information such as the target distribution cabinet's device identifier, the third predicted fault type, the time the third predicted fault type was acquired, and the fourth predicted probability corresponding to the third predicted fault type. This generates abnormality warning information corresponding to the target distribution cabinet, prompting supervisors to implement appropriate manual inspections or fault response plans, thereby helping to ensure the safe and stable operation of the power system. The specific content of the information template can be customized by the implementer based on actual needs.

[0110] It should be noted that if the predicted fault type is no fault, no abnormal warning information will be generated.

[0111] As mentioned above, when the predicted fault type is not predicted, the first preset value is used as the sampling time interval between adjacent preset time points, and current data is collected at a lower frequency to save equipment power consumption and reduce the cost of data collection and transmission. When it is difficult to accurately predict the fault type based on the first prediction sequence, the sampling time interval is adjusted, and current data is collected at a higher frequency in the second time period. If the accuracy of the fault type prediction does not improve after the sampling time interval is adjusted, the collection of the operating status information of the target distribution cabinet is started in the third time period, and then subsequent fault prediction is performed, thereby performing data collection and fault prediction in a staged manner. Compared with the method of continuously collecting data at a high frequency in the prior art, it can greatly reduce the cost of abnormal warning while ensuring the accuracy of fault prediction, especially in the scenario where the data transmission bandwidth of the server is limited, it can enable the server to support current abnormality monitoring and early warning of more distribution cabinets.

[0112] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any form. Although the present invention has been disclosed as above in terms of preferred embodiments, they are not intended to limit the present invention. Any technician familiar with this profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for early warning of abnormal current in distribution cabinets based on big data, characterized in that: The method comprises the following steps: S1, based on a first acquisition sequence formed by current acquisition values ​​corresponding to each first preset time point of the target distribution cabinet in the first time period, predicting a first prediction sequence formed by first current prediction values ​​corresponding to each second preset time point of the target distribution cabinet in a fourth time period, wherein the time interval between adjacent first preset time points is a first preset value, the time interval between adjacent second preset time points is a second preset value, and the first preset value is an integer multiple of the second preset value; S2: If the first prediction sequence satisfies a first preset condition, sampling according to a time interval corresponding to a second preset value to obtain a second acquisition sequence formed by current acquisition values ​​of the target power distribution cabinet corresponding to each second preset time point in the second time period; S3, sampling the second acquisition sequence according to the time interval corresponding to the first preset value, to obtain a third acquisition sequence formed by the current acquisition values ​​of the target power distribution cabinet corresponding to each first preset time point in the second time period; S4, if the second acquisition sequence and the third acquisition sequence meet the second preset condition, obtaining a fourth acquisition matrix formed by the current acquisition values ​​and operating status information corresponding to each second preset time point of the target power distribution cabinet in the third time period, wherein S4 includes the following steps: S41, inputting the second acquisition sequence into a trained time series prediction model to obtain second current prediction values ​​of the target distribution cabinet corresponding to each second preset time point in the fourth time period; S42, combining the second current prediction values ​​corresponding to the second preset time points of the target power distribution cabinet in the fourth time period in chronological order to obtain a second prediction sequence; S43, obtaining second prediction probabilities corresponding to M fault types respectively according to the second prediction sequence; S44, inputting the third acquisition sequence into the trained time series prediction model to obtain third current prediction values ​​of the target distribution cabinet corresponding to each second preset time point in the fourth time period; S45, splicing the third current prediction values ​​of the target power distribution cabinet corresponding to each second preset time point in the fourth time period in chronological order to obtain a third prediction sequence; S46, obtaining third prediction probabilities corresponding to the M fault types respectively according to the third prediction sequence; S47, judging whether the second acquisition sequence and the third acquisition sequence meet a second preset condition based on the second predicted probabilities and the third predicted probabilities corresponding to the M fault types respectively; S5: Predicting, based on the fourth acquisition matrix, fourth predicted probabilities corresponding to the target power distribution cabinet for M fault types; S6. Obtain abnormal warning information corresponding to the target power distribution cabinet according to the M fourth prediction probabilities.

2. The method for early warning of abnormal current in distribution cabinets based on big data according to claim 1 is characterized in that: S1 includes the following steps: S11, starting from the starting time point of the first time period, collecting current collection values ​​of the target power distribution cabinet corresponding to each first preset time point in the first time period; S12, splicing the current collection values ​​corresponding to each first preset time point of the target power distribution cabinet in the first time period in chronological order to obtain the first collection sequence; S13: inputting the first acquisition sequence into a trained time series prediction model to obtain first current prediction values ​​of the target power distribution cabinet corresponding to each second preset time point in the fourth time period; S14: splicing the first current prediction values ​​of the target power distribution cabinet corresponding to each second preset time point in the fourth time period in chronological order to obtain the first prediction sequence.

3. The method for early warning of abnormal current in distribution cabinets based on big data according to claim 2 is characterized in that: S2 includes the following steps: S21, splicing the first prediction sequence and a preset zero vector matrix to obtain a first splicing result; S22: Input the first splicing result into a trained fault prediction model to predict first prediction probabilities corresponding to the target power distribution cabinet for M fault types; S23: If the maximum value of the M first prediction probabilities is less than a preset probability threshold, it is determined that the first prediction sequence meets a first preset condition.

4. The method for early warning of abnormal current in a distribution cabinet based on big data according to claim 3 is characterized in that: S2 also includes the following steps: S24, starting from the start time point of the second time period, collecting current collection values ​​of the target power distribution cabinet corresponding to each second preset time point in the second time period; S25 , splicing the current collection values ​​of the target power distribution cabinet corresponding to each second preset time point in the second time period in chronological order to obtain the second collection sequence.

5. The method for early warning of abnormal current in a distribution cabinet based on big data according to claim 1, characterized in that: S43 includes the following steps: S431, splicing the second prediction sequence and a preset zero vector matrix to obtain a second splicing result; S432: Input the second splicing result into a trained fault prediction model to predict and obtain second prediction probabilities corresponding to the target power distribution cabinet for M fault types.

6. The method for early warning of abnormal current in a distribution cabinet based on big data according to claim 5 is characterized in that: S46 includes the following steps: S461, splicing the third prediction sequence and the preset zero vector matrix to obtain a third splicing result; S462: Input the third concatenation result into the trained fault prediction model to predict and obtain third prediction probabilities corresponding to the target distribution cabinet for M fault types.

7. The method for early warning of abnormal current in a distribution cabinet based on big data according to claim 6, characterized in that: S47 includes the following steps: S471: Determine the fault type corresponding to the maximum second predicted probability of the M fault types as the first predicted fault type corresponding to the target power distribution cabinet; S472: Determine the fault type corresponding to the maximum third predicted probability among the M fault types as the second predicted fault type corresponding to the target power distribution cabinet; S473, subtracting the maximum second prediction probability from the maximum third prediction probability to determine the prediction probability difference; S474: If the first predicted fault type and the second predicted fault type are inconsistent, or the predicted probability difference is less than a preset probability difference threshold, determine that the second acquisition sequence and the third acquisition sequence meet a second preset condition.

8. The method for early warning of abnormal current in a distribution cabinet based on big data according to claim 6 is characterized in that: S4 also includes the following steps: S401, starting from the start time point of the third time period, collecting current collection values ​​corresponding to each second preset time point in the third time period of the target distribution cabinet, and corresponding operating status information, wherein each operating status information includes status data corresponding to I operating status types, where I is a positive integer; S402: splicing the current collection values ​​corresponding to each second preset time point of the target power distribution cabinet in the third time period in chronological order to obtain a fourth collection sequence; S403, performing vectorization processing on each piece of running status data to obtain a state vector corresponding to each piece of running status data; S404, determining the position of each state vector in the state matrix according to the chronological order and the preset order of the operating state type; S405, obtaining a state matrix according to the position of each state vector in the state matrix; S406: Concatenate the fourth acquisition sequence and the state matrix to obtain the fourth acquisition matrix.

9. The method for early warning of abnormal current in a distribution cabinet based on big data according to claim 1, characterized in that: S6 includes the following steps: S61, determining the fault type corresponding to the largest fourth predicted probability as the third predicted fault type corresponding to the target power distribution cabinet; S62: Generate the abnormal warning information according to the third predicted fault type.

Images