A method and system for monitoring residual current of an alternating current system for a station

Abstract

The application discloses a kind of station with alternating current system residual current monitoring method and system, the method includes: cloud server receives the current signal sent by monitoring field mobile terminal, the current signal includes the current signal of three phase lines and one neutral line;Based on the deviation change trend of historical measurement results of residual current sensing device, correct current signal;Based on the current signal of four lines after correction, vector synthesis is carried out, and residual current signal is obtained;Based on current residual current signal, extract feature vector, based on feature vector analysis whether fault occurs and the fault type that occurs.The application uses mobile terminal and cloud to jointly realize residual current monitoring, improve the accuracy and efficiency of residual current monitoring analysis, uses the fault analysis of residual current signal by cloud, realizes the monitoring of line electrical safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of alternating current system safety monitoring, and particularly relates to a method and system for monitoring residual current of a station alternating current system. BACKGROUND

[0002] The station alternating current system provides power supply for primary and secondary devices of a substation, and is an indispensable link for ensuring reliable power supply of the substation. Abnormal conditions will directly affect the normal use of the running devices. The existence of residual current may cause various problems.

[0003] 1. Electrical fire: due to the existence of residual current, electrical systems may experience arc discharge, electric sparks and other phenomena, which may lead to electrical fires.

[0004] 2. Device damage: persistent residual current may cause internal heating of the device, accelerating the aging and even damaging the device.

[0005] 3. System performance degradation: due to the existence of residual current, additional energy consumption and voltage drop may occur in the system, affecting the performance of the system.

[0006] 4. Threat to personal safety: if the device shell is electrified or someone touches a damaged electrical device, an electric shock accident may occur, threatening personal safety.

[0007] The residual current fault risk of the station alternating current system will bring major safety hazards to the station alternating current power supply, so it is very important to monitor the residual current of the station alternating current system. In the prior art, due to factors such as measurement error and synthesis error, there is a deviation between the residual current monitoring and the actual value, which further leads to misjudgment of the residual current fault. SUMMARY

[0008] In view of the above problems existing in the prior art, the present application provides a method and system for monitoring residual current of a station alternating current system, and the technical solution is as follows:

[0009] In the first aspect, a method for monitoring residual current of a station alternating current system is provided, and the method comprises the following steps:

[0010] The cloud server receives the current signal sent by the monitoring field mobile terminal, and the current signal comprises current signals of three phase lines and one neutral line;

[0011] The current signal correction model includes an encoder, a decoder and a mapping prediction model. The encoder is used to extract current signal feature information. The decoder is used to reconstruct the current signal based on the current signal feature information. The mapping prediction model is used to map the current signal feature information extracted by the encoder to current signal feature information after removing the measurement deviation, and reconstruct the current signal after removing the measurement deviation by using the decoder.

[0012] The four corrected line current signals are vector synthesized to obtain a residual current signal.

[0013] Based on the current residual current signal, a feature vector is extracted, and whether a fault occurs and the type of the fault is analyzed based on the feature vector.

[0014] In some embodiments, the field mobile terminal is wirelessly or wiredly connected to the four residual current sensing devices. The communication method between the field mobile terminal and the residual current sensing devices includes:

[0015] The field mobile terminal sends a sampling synchronization signal and a preset sampling duration to the four residual current sensing devices.

[0016] The residual current sensing device starts an analog-to-digital converter to collect the current signal input by the alternating current transformer based on the received sampling synchronization signal, and obtains the collected current signal after the preset sampling duration.

[0017] The residual current sensing device sends the collected current signal to the field mobile terminal.

[0018] In some embodiments, the field mobile terminal sends a sampling synchronization signal and a preset sampling duration to the residual current sensing device, which includes:

[0019] The field mobile terminal sends a same first reference signal to the four residual current sensing devices.

[0020] The residual current sensing device generates a first reference signal reception completion time after receiving the first reference signal.

[0021] The residual current sensing device sends the first reference signal reception completion time to the field mobile terminal to determine the reception delay error of the four residual current sensing devices.

[0022] The field mobile terminal sends a sampling synchronization signal, a preset sampling duration and a reception delay duration to the four residual current sensing devices.

[0023] In some embodiments, the residual current sensing device sends the collected current signal to the field mobile terminal, including:

[0024] The four residual current sensing devices splice the preset second reference signal with the collected current signal, and send the spliced signal to the field mobile terminal;

[0025] The residual current signal is obtained by vector synthesis based on the corrected four line current signals, including aligning the four line current signals, and the alignment includes:

[0026] Based on the four line current signals, the preset second reference signal in the four line current signals is identified, and the alignment is performed based on the preset second reference signal in the four line current signals, so as to determine the alignment of the current signals collected by the four residual current sensing devices.

[0027] In some embodiments, the preset current signal correction model includes:

[0028] The first encoder is used to input the current signal and obtain the key features of the input signal;

[0029] The first decoder is connected with the first output of the first encoder and the first input of the first decoder; and is used to generate the current signal based on the input key features;

[0030] The first mapping prediction model is connected with the second output of the first encoder and the first input of the first mapping prediction model; and is used to obtain the corresponding mapping key features based on the input key features;

[0031] The method for correcting the current signal by the current signal correction model includes:

[0032] Based on the current collected current signal of the current residual current sensing device, the first encoder is input, and the first encoder outputs the third key features;

[0033] The third key features are input to the first mapping prediction model, and the first mapping prediction model outputs the fourth key features;

[0034] The fourth key features are input to the first decoder, and the first decoder outputs the reconstructed current signal, i.e. the corrected current signal.

[0035] In some embodiments, the method for obtaining the preset current signal correction model includes:

[0036] Based on the current signal collected by the residual current sensing device, the preset encoder network is input to obtain the encoding features, and the encoding features are input to the preset decoder network to obtain the decoding features, i.e. the reconstructed current signal;

[0037] constructing an error loss function based on the current signal input to the encoder network and the reconstructed current signal output by the decoder network, iteratively training the encoder network and the decoder network based on the error loss to obtain a trained first encoder and a trained first decoder;

[0038] inputting a measurement result of a historical acquisition signal of the residual current sensing device into the first encoder to obtain a first key feature, and inputting an actual current signal of the historical acquisition signal into the first encoder to obtain a second key feature;

[0039] inputting the first key feature into a preset mapping prediction model to obtain an output key feature, constructing an error loss function based on the output key feature and the second key feature, and iteratively training the mapping prediction model based on the error loss to obtain a trained first mapping prediction model.

[0040] In some embodiments, the inputting the fourth key feature into the first decoder includes:

[0041] taking the current acquisition current signal of the current residual current sensing device as an additional reference signal for the reconstructed current signal output by the first decoder;

[0042] inputting the additional reference information and the fourth key feature into the first decoder to obtain the reconstructed current signal, i.e., the corrected current signal;

[0043] The first decoder includes a self-attention unit and a cross-attention unit, the additional reference signal is input into the self-attention unit, the key matrix K, the query matrix Q, and the value matrix V of the self-attention unit are calculated based on the additional reference signal to obtain additional reference signal feature information, the cross-attention unit is input based on the additional reference signal feature information and the fourth key feature to obtain a decoding signal taking the additional reference signal as a reference, wherein the Q matrix of the cross-attention unit is calculated based on the fourth key feature as input, and the K and V matrices of the cross-attention unit are calculated based on the additional reference signal feature information as input.

[0044] In some embodiments, the extracting a feature vector based on the current residual current signal includes:

[0045] extracting a time domain feature based on the current residual current signal, and extracting a frequency domain feature based on frequency distribution data obtained by time-frequency transformation of the current residual current signal;

[0046] Based on time domain features and frequency domain features, an integrated model is used to identify the type of fault occurring; the integrated model includes a plurality of fault type identification models that have been trained.

[0047] In a second aspect, a residual current monitoring system for an AC system of a station is provided, and the system comprises:

[0048] A signal receiving unit is configured to receive a current signal transmitted by a mobile terminal in a monitoring field, wherein the current signal comprises current signals of three phase lines and a neutral line.

[0049] A signal correction unit is configured to correct the received current signal based on a historical measurement deviation trend of a residual current sensing device, using a preset current signal correction model, wherein the current signal correction model comprises an encoder, a decoder and a mapping prediction model, the encoder is configured to extract feature information of the current signal, the decoder is configured to reconstruct the current signal based on the feature information of the current signal, and the mapping prediction model is configured to map the feature information of the current signal extracted by the encoder to feature information of the current signal after removing the measurement deviation, and reconstruct the current signal after removing the measurement deviation using the decoder.

[0050] A signal synthesis unit is configured to perform vector synthesis based on the corrected current signals of the four lines to obtain a residual current signal.

[0051] A fault analysis unit is configured to extract a feature vector based on the current residual current signal, and analyze whether a fault occurs and the type of the fault based on the feature vector.

[0052] In a third aspect, a computer readable storage medium is provided, which stores computer instructions, and the instructions are executed by a processor to implement the steps of the residual current monitoring method for an AC system of a station according to the first aspect.

[0053] The residual current monitoring method and system for an AC system of a station have the following beneficial effects: the present application takes into account the measurement errors of different residual current sensing devices and the errors caused by different line signal synthesis, corrects the collected current signal first, then performs vector synthesis, obtains a residual current signal based on the vector synthesis result, analyzes whether a fault occurs based on the residual current signal, and realizes residual current monitoring by combining a mobile terminal and a cloud, improves the accuracy and efficiency of residual current monitoring, and realizes the monitoring of line electrical safety by analyzing the fault of the residual current signal. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 FIG. 1 is a flowchart of a residual current monitoring method for an AC system of a station according to an embodiment of the present application;

[0055] Figure 2 is a structural schematic diagram of a current signal correction model in the embodiment of the application;

[0056] Figure 3 is a structural schematic diagram of a residual current monitoring system of an AC system for a substation in the embodiment of the application. DETAILED DESCRIPTION

[0057] It should be understood that the specific embodiments described herein are merely illustrative of the application and do not limit the application.

[0058] Referring to Figure 1 The embodiment of the application provides a residual current monitoring method of an AC system for a substation, which is applied to a cloud server and includes the following steps:

[0059] Step 1: The cloud server receives a current signal sent by a monitoring field mobile terminal, wherein the current signal includes current signals of three phase lines and one neutral line.

[0060] Step 2: Based on a historical measurement result deviation change trend of a residual current sensing device, a preset current signal correction model is used to correct the received current signal, wherein the current signal correction model includes an encoder, a decoder and a mapping prediction model, the encoder is used to extract current signal feature information, the decoder is used to reconstruct the current signal based on the current signal feature information, and the mapping prediction model is used to map the current signal feature information extracted by the encoder to current signal feature information after removing measurement deviation, and the current signal feature information after removing measurement deviation is used to reconstruct the current signal after removing measurement deviation by using the decoder.

[0061] Step 3: Vector synthesis is performed based on the corrected current signals of the four lines to obtain a residual current signal.

[0062] Step 4: Based on the current residual current signal, a feature vector is extracted, and whether a fault occurs and a fault type are analyzed based on the feature vector.

[0063] In the embodiment of the application, the field mobile terminal is used to collect current signals of three phase lines and one neutral line of a line, considering that different residual current sensing devices have their own measurement errors, the collected current signals are corrected first, then vector synthesis is performed, a residual current signal is obtained based on the vector synthesis result, and whether a fault occurs is analyzed based on the residual current signal. The application uses a mobile terminal and a cloud to jointly realize residual current monitoring, improves the accuracy and efficiency of residual current monitoring, and realizes monitoring of electrical safety of a line by using fault analysis of the residual current signal.

[0064] In an embodiment, the field mobile terminal is wirelessly or wiredly connected with four residual current sensing devices, and a communication method between the field mobile terminal and the residual current sensing devices comprises the following steps:

[0065] In step 101, the field mobile terminal sends a sampling synchronization signal and a preset sampling duration to the four residual current sensing devices;

[0066] In step 102, the residual current sensing devices start to collect the current signals inputted by the alternating current transformers based on the received sampling synchronization signal, and obtain the collected current signals after the preset sampling duration;

[0067] In step 103, the residual current sensing devices send the collected current signals to the field mobile terminal.

[0068] In the embodiment, the field mobile terminal and the residual current sensing devices have the ability to communicate with each other. For example, the field mobile terminal sends a synchronization sampling signal in a broadcast mode through 470MHz wireless communication, the residual current sensing devices collect current signals based on the synchronization sampling signal, and can send the collected current signals to the field mobile terminal. Of course, the mobile terminal and the residual current sensing devices can also use Ethernet communication, for example, send information by using UDP protocol. Each residual current sensing device has a microprocessor and a display screen, can display the measured current signal, can calibrate the sensor to full range, and can mark which phase line the measured current is. Each mobile terminal has a microprocessor and a display screen, can display a plurality of residual current sensing devices which have established a communication connection, and can view the stored measured current and its curve diagram. Each residual current sensing device can adopt a detachable ring-shaped sensing structure design, which is convenient to use and has no secondary open circuit danger. In the embodiment, the mobile terminal and the residual current sensing device are designed in a split type, which can adapt to complex field conditions and improve the application scene applicability.

[0069] In an embodiment, the step 101 of sending a sampling synchronization signal and a preset sampling duration to the residual current sensing devices by the field mobile terminal comprises the following steps:

[0070] In step 1011, the field mobile terminal sends a same first reference signal to the four residual current sensing devices;

[0071] In step 1012, the residual current sensing devices generate a first reference signal receiving completion time after receiving the first reference signal;

[0072] In step 1013, the residual current sensing devices send the first reference signal receiving completion time to the field mobile terminal, and determine the receiving delay error of the four residual current sensing devices.

[0073] Step 1014, the on-site mobile terminal sends a sampling synchronization signal, a preset sampling duration, and a receiving delay duration to the four remaining current sensing devices.

[0074] In the embodiment of the application, the communication time error between each remaining current sensing device and the mobile terminal is measured by using a known first reference signal. It can be understood that the difference in the completion time of receiving the first reference signal by each remaining current sensing device is used, one of which is selected as a standard with the latest completion time, to determine the delay error of the receiving time of each remaining current sensing device and the standard time. After compensating for the delay error, the current signals of each line can be synchronously collected, ensuring that each remaining current sensing device can collect the current signals of each line at the same time.

[0075] In one embodiment, step 102, the remaining current sensing device starts an analog-to-digital converter to collect the current signal input by the alternating current transformer based on the received sampling synchronization signal, and acquires the collected current signal after the preset sampling duration, including: the remaining current sensing device starts an analog-to-digital converter to collect the signal input by the alternating current transformer after the receiving delay duration based on the received sampling synchronization signal, and acquires the collected current signal after the preset sampling duration.

[0076] In the embodiment of the application, it can be understood that each remaining current sensing device can start an analog-to-digital converter to collect the signal input by the alternating current transformer after the receiving delay duration + the fixed buffering duration after receiving the sampling synchronization signal. The fixed buffering duration corresponding to each remaining current sensing device is the same, ensuring that each remaining current sensing device can simultaneously start collecting the current signals of each line. The preset sampling duration can be the duration of the signal that needs to be collected, for example, 200 milliseconds.

[0077] In one embodiment, step 103, the remaining current sensing device sends the collected current signal to the on-site mobile terminal, including:

[0078] Step 1031, the four remaining current sensing devices splice the preset second reference signal with the collected current signal, and send the spliced signal to the on-site mobile terminal.

[0079] The vector synthesis based on the corrected current signals of the four lines to obtain the residual current signal, including: aligning the current signals of the four lines, the alignment including:

[0080] In step 301, based on the four-line current signals, a preset second reference signal in the four-line current signals is identified, and alignment is performed based on the preset second reference signal in the four-line current signals, so as to determine the alignment of the current signals collected by the four remaining current sensing devices.

[0081] In the embodiment of the application, the preset second reference signal is used to realize the alignment of the four-line current signals. It can be understood that, by identifying the spliced signal and identifying the second reference signal therein, the alignment of the actually collected current signals can be performed based on the second reference signal of the signals collected on each line. It should be noted that the second reference signal used by each remaining current sensing device during splicing is the same.

[0082] Referring to Figure 2 In an embodiment, the preset current signal correction model includes:

[0083] The first encoder 201 is configured to input the current signal and obtain the key features of the input signal.

[0084] The first decoder 202 is connected with the first output of the first encoder and the first input of the first decoder. The first decoder is configured to generate the current signal based on the input key features.

[0085] The first mapping prediction model 203 is connected with the second output of the first encoder and the first input of the first mapping prediction model. The first mapping prediction model is configured to obtain the corresponding mapping key features based on the input key features.

[0086] The current signal correction model corrects the method of the current signal, which includes:

[0087] In step 21, the current collected current signal of the current remaining current sensing device is input into the first encoder, and the first encoder outputs the third key features.

[0088] In step 22, the third key features are input into the first mapping prediction model, and the first mapping prediction model outputs the fourth key features. The fourth key features represent the actual key features of the original current signal of the current collected current signal.

[0089] In step 23, the fourth key features are input into the first decoder, and the first decoder outputs the reconstructed current signal, i.e., the corrected current signal.

[0090] In an embodiment, the method for obtaining the preset current signal correction model includes:

[0091] Step (A1), based on the current signal collected by the residual current sensing device, input to the preset encoder network, get the encoding feature, input the encoding feature to the preset decoder network, get the decoding feature, i.e. the reconstructed current signal;

[0092] Step (A2), based on the current signal input to the encoder network and the reconstructed current signal output by the decoder network, construct the error loss function, based on the error loss, iteratively train the encoder network and the decoder network, get the trained first encoder and first decoder, in the iterative training process, the reconstructed current signal output by the decoder network gradually approaches the current signal input to the encoder;

[0093] Step (A3), based on the measurement result of the historical acquisition signal of the residual current sensing device (including the current signal with measurement error), input to the first encoder, get the first key feature, based on the actual current signal of the historical acquisition signal (the current signal without measurement error), input to the first encoder, get the second key feature;

[0094] Step (A4), based on the first key feature input to the preset mapping prediction model, get the output key feature, based on the output key feature and the second key feature, construct the error loss function, based on the error loss, iteratively train the mapping prediction model, get the trained first mapping prediction model.

[0095] In the embodiments of the present application, considering that the residual current signal contains multiple components, and the measurement interference factors existing in the residual current sensor device have different interference effects on different components, in the embodiments of the present application, the measurement results of the historical signals and the actual current signal data are used to train the first mapping prediction model to learn the influence of the measurement interference factors existing in the residual current sensor device on the key features, which are used to represent the influence of the measurement interference factors existing in the residual current sensor device on different components. In the embodiments of the present application, the error loss function can be implemented based on multiple functions representing the degree of error (for example, mean square error loss function, etc.), and the error loss function that is optimal for the training process can be selected. In the embodiments of the present application, the first encoder, the first mapping prediction model and the first decoder are combined, and the key features extracted by the first encoder can be used to fully express the key information of the current signal. These key features can reconstruct the current signal. Based on the error of these key features in the measurement signal and the actual signal, the measurement error of a residual current sensor device when collecting various current signals can be learned, and the correction of the current signal facing the individual characteristics of the residual current sensor device is realized. The first encoder is used to extract the features of the current signal, and can be implemented by using various models capable of feature extraction, such as convolutional neural network (CNN), gated recurrent network (GRU), long short-term memory (LSTM) model, etc. The decoder is used to generate the time series current signal based on the key features, and can also use a convolutional neural network model. The difference lies in that the data processing process of the encoder and the data processing process of the decoder are inverse. For example, the first encoder and the first decoder can also be jointly implemented by using a model, such as the unet model. It should be noted that if the Unet network structure is used, the output of the first encoder is directly input to the input of the first decoder during the training of the current signal correction model, and the output of the first encoder is connected to the mapping prediction model during the use of the current signal correction model, and the output of the mapping prediction model is input to the input of the first decoder.

[0096] In an embodiment, the mapping prediction model can adopt a multi-level mapping unit joint structure with multiple mapping units predicting step by step. Each level of mapping unit further predicts the measurement deviation based on the measurement deviation prediction error of the previous level. The output of the last mapping unit of the multiple mapping units is the measurement deviation prediction result of the device. The multiple mapping units learn the contribution of different measurement influencing factors of the device to the measurement error, for example, the data processing steps performed by the mapping prediction model include:

[0097] (A101) Obtain the output features of the measurement data (current signal containing measurement error) of the device through the first encoder, denoted as the first features, and obtain the output features of the true current signal (current signal without measurement error) of the measurement data of the device through the first encoder, denoted as the second features;

[0098] (A102) Training a first mapping unit to learn device physical property mapping features based on the first features as input and the second features as output, the first mapping unit adopts a radial basis function network to learn the error caused by the physical property of the device itself, the initial center of the radial basis function is determined based on the clustering center of the error data of the first features and the second features, and is iteratively optimized during the training process; after the training is completed, the difference between the input and the output of the first mapping unit is taken as the first mapping feature;

[0099] (A103) Training a second mapping unit to learn external factor mapping features based on the difference between the first features and the first mapping features as the first correction features, jointly taking the environmental parameters (temperature, humidity, etc.) and the device state parameters (current, voltage, etc.) as input, and the second features as output, the second mapping unit adopts a three-layer fully connected layer; the difference between the first correction features of the second mapping unit and the output of the second mapping unit is taken as the second mapping feature;

[0100] (A104) Training a third mapping unit to learn time sequence mapping features based on the difference between the first features and the first mapping features and the second mapping features as the second correction features, jointly taking the time sequence of the second correction features, the environmental parameters, and the device state parameters as input, and the second features as output, the third mapping unit adopts a time sequence prediction model, which can adopt an LSTM model; the difference between the second correction features and the output of the third mapping unit is taken as the third mapping feature;

[0101] (A105) Taking the output of the third mapping unit as the output of the mapping prediction model.

[0102] By using this multi-level mapping unit mode, each mapping unit can learn to predict a type of error separately, avoiding interference between different types of errors, improving the prediction accuracy of each type of error, and gradually excluding the contribution of the measurement error influencing factors of the previous type of measurement error to the measurement error.

[0103] In the training process, the training samples adopted by the first mapping unit, the second mapping unit and the third mapping unit can be selected according to the needs of appropriate samples and corresponding supervision label data. For example, the training of the first mapping unit can be based on the device measurement data in a short time (to avoid the change of the physical characteristics of the device over time) under the condition of stable environment and stable device state as sample data, and the real current signal without measurement error as label data, to train the first mapping unit. The loss function is constructed based on the output of the first mapping unit and the error of the second feature, and the first mapping unit learns the measurement error caused by the physical characteristics of the device. The training of the second mapping unit can use the first corrected feature, the environment parameter and the device state parameter at different time points as input, and the output of the second mapping unit and the error of the second feature to construct a loss function to train the second mapping unit to learn the measurement error caused by the environment and device state parameters. The training of the third mapping unit can use the time sequence of the second corrected feature, the environment parameter and the device state parameter as input, and the output of the third mapping unit and the error of the second feature to construct a loss function to train the third mapping unit to learn the measurement error caused by the time sequence factor.

[0104] In another embodiment, the mapping prediction model performs feature decomposition and feature conversion to the corresponding feature space based on the output features of the encoder, so that the current signal feature information after removing the measurement deviation and the measurement deviation feature information are converted to two different feature spaces to obtain the current signal feature information after removing the measurement deviation. The mapping prediction model includes a feature decomposition layer and a feedforward neural network layer connected in sequence, and the feature decomposition layer includes two parallel and corresponding position complementary attention weight layers, respectively denoted as a first feature decomposition layer and a second feature decomposition layer. Further, a classification layer can be arranged at the output of the feature decomposition layer and the feedforward neural network layer, respectively, for identifying whether the output feature of the previous layer is a feature containing measurement error information.

[0105] The training process of the mapping prediction model includes:

[0106] (A201) Obtain the output feature of the first encoder of the measurement data (current signal containing measurement error) of the device, denoted as the first feature, and the output feature of the first encoder of the real current signal (current signal without measurement error) of the measurement data of the device, denoted as the second feature;

[0107] (A202) input the first feature into the mapping prediction model to be trained, wherein the mapping prediction model comprises a feature decomposition layer and a feedforward neural network layer, the feature decomposition layer comprises two parallel and position-complementary attention weight layers, respectively denoted as a first feature decomposition layer and a second feature decomposition layer; the first feature is input into the mapping prediction model, and the first real feature and the first error feature are obtained through the two attention weight layers of the feature decomposition layer; the first real feature and the first error feature pass through the feedforward neural network layer to obtain the first transformed feature and the first error transformed feature, respectively;

[0108] (A203) input the second feature into the mapping prediction model to obtain the second real feature, the second error feature, the second transformed feature and the second error transformed feature in sequence;

[0109] (A204) determine the error loss function of the mapping prediction model based on the similarity of the first real feature and the second real feature, the difference between the first error feature and the second error feature and the error of the first error feature, the similarity error of the first transformed feature and the second transformed feature, the difference between the first error transformed feature and the second error transformed feature and the error of the first error transformed feature, and train the mapping prediction model based on the error loss function.

[0110] After the mapping prediction model is trained, in actual use, the following steps are included: obtaining the output feature of the first encoder as the third feature; inputting the third feature into the mapping prediction model to obtain the third real feature and the third error feature through the two attention weight layers of the feature decomposition layer; and inputting the third real feature into the decoder to generate the current signal.

[0111] Specifically, the weight parameters of the same position of the two attention weight layers of the feature decomposition layer are complementary, for example, the weight parameter of the jth position of the A attention weight layer is 0.3, and the weight parameter of the jth position of the B attention weight layer is 0.7, and the weight parameters of the same position of the two attention weight layers add up to 1.

[0112] Further, the first features and the second features corresponding to the batch of device measurement data can be simultaneously mapped to the training of the mapping prediction model, rather than inputting the first features and the second features of one measurement data at a time, and then training the model parameters by using the error loss function. In this way, the batch data is used to train the model, and the training efficiency of the model is improved. In this case, for example, the first features and the second features of 10 measurement data are used as one batch of data to perform one iteration update of the model parameters. Further, at this time, the mapping prediction model further includes a classification layer after the feature decomposition layer and the feedforward neural network layer, respectively, for identifying whether the output features are features containing measurement deviation. The classification layer can use a softmax function. The error loss function used by the mapping prediction model in the training further includes a third error, which includes the category consistency of all the first real features and the second real features corresponding to the plurality of measurement data, the category consistency of all the first error features, the category consistency of all the first transformed features and the second transformed features, and the category consistency of all the first error transformed features.

[0113] The mapping prediction model in the embodiments of the present application uses two parallel and position-complementary attention weight layers to decompose the current signal features, to obtain real features without measurement error and measurement error features, and projects and transforms the real features and the measurement error features to different spatial features through a joint feedforward neural network layer, so that the mapping prediction model can accurately distinguish between features without measurement error and features with measurement error, and can decompose real features containing various measurement errors of different degrees and different types. The second features without measurement error and the first features with measurement error are simultaneously used as the training input of the mapping prediction model, and the differences and hidden combination characteristics of the two are learned in the model. It can be understood that the second features without measurement error have a reference comparison function in the training process.

[0114] Of course, in the present application, the training processes of the encoder, the decoder and the mapping prediction model can also be jointly trained, as follows. The method for obtaining the preset current signal correction model comprises:

[0115] Step (B1), based on the current signal collected by the residual current sensing device, the current signal is input to the preset encoder network to obtain the encoded features, and the encoded features are input to the preset decoder network to obtain the decoded features, i.e. the reconstructed current signal;

[0116] Step (B2), based on the current signal input to the encoder network and the reconstructed current signal output by the decoder network, an error loss function is constructed, and the encoder network and the decoder network are iteratively trained based on the error loss to obtain a preliminarily trained preliminary encoder and a preliminarily trained preliminary decoder;

[0117] Step (B3), based on the measurement results of the historical acquisition signals of the residual current sensing device, sequentially input to the preliminary encoder, the preset mapping prediction model, the preliminary decoder, obtain the final output of the preliminary decoder, based on the error between the final output of the preliminary decoder and the original current signal of the acquisition signal, the error between the output result of the preset mapping prediction model and the result obtained by inputting the original current signal of the acquisition signal to the preliminary encoder, joint construction of error loss function, iterative training of preliminary encoder, preset mapping prediction model, preliminary decoder, to obtain the first trained encoder, first mapping prediction model, first decoder.

[0118] In the embodiments of the application, the encoder and the decoder are pre-trained first, and then the preliminary trained encoder and the decoder are fine-tuned simultaneously while training the mapping prediction model, and the joint training of the preliminary encoder, the preset mapping prediction model and the preliminary decoder is performed, which can improve the accuracy of current signal correction.

[0119] In one embodiment, the above step 23, based on the fourth key feature input to the first decoder, the first decoder outputs the reconstructed current signal, i.e. the corrected current signal, comprising:

[0120] Step 231, taking the current acquisition current signal of the current residual current sensing device as an additional reference signal for the first decoder to output the reconstructed current signal;

[0121] Step 232, input the additional reference information and the fourth key feature to the first decoder, and the first decoder outputs the reconstructed current signal, i.e. the corrected current signal;

[0122] The first decoder includes a self-attention unit and a cross-attention unit, the additional reference signal is input to the self-attention unit, the key matrix K (Key), the query matrix Q (Query) and the value matrix V (Value) of the self-attention unit are calculated based on the additional reference signal to obtain the additional reference signal feature information; the cross-attention unit is input based on the additional reference signal feature information and the fourth key feature to obtain the decoding signal taking the additional reference signal as the reference, wherein the Q matrix of the cross-attention unit is calculated based on the fourth key feature as the input, and the K and V matrices of the cross-attention unit are calculated based on the additional reference signal feature information as the input.

[0123] In the embodiments of the present application, the current collected current signal of the current residual current sensing device is taken as a reference signal to assist the signal generation process of the decoder, thereby improving the accuracy and efficiency of the reconstructed current signal output by the first decoder. The K and V matrices of the cross-attention unit are calculated using the additional reference signal feature information, and the Q matrix of the cross-attention unit is calculated using the fourth key feature, thereby realizing the guidance of the additional reference signal to the signal generation of the decoder. It can be understood that the first decoder can further have a feedforward neural network layer and an output layer for outputting the generated signal. Specifically, the output of the cross-attention unit is input to the feedforward neural network layer after residual addition and normalization processing, the output of the feedforward neural network layer is processed by the fully connected layer after residual addition and normalization processing, and then the processed signal is output by the output layer. an output layer for outputting the generated signal. Specifically, the output of the cross-attention unit is input to the feedforward neural network layer after residual addition and normalization processing, the output of the feedforward neural network layer is processed by the fully connected layer after residual addition and normalization processing, and then the processed signal is output by the output layer.

[0124] It can be understood that in the self-attention unit and the cross-attention unit, the input features are decomposed into K, Q, and V matrices, wherein, , is the dimension of the K matrix, wherein the K, Q, and V matrices can be obtained based on the same input and corresponding weight matrices (self-attention) or different inputs and corresponding weight matrices (cross-attention), wherein each weight matrix used to calculate the K, Q, and V matrices is different and is obtained by learning.

[0125] In one embodiment, in the above step 4, based on the current residual current signal, a feature vector is extracted, and whether a fault occurs and the type of the fault are analyzed based on the feature vector, including:

[0126] Step 401: Based on the current residual current signal, time domain features are extracted; frequency distribution data is obtained by time-frequency transformation of the current residual current signal, and frequency domain features are extracted based on the frequency distribution data; the time domain features include mean value, variance, extreme value, etc., and can also include other features representing key information of the waveform, such as steepness; the frequency domain features include the amplitudes of the fundamental wave and each harmonic signal and the proportion of the harmonic signal;

[0127] Step 402: Based on the time domain features and the frequency domain features, an integrated model is used to identify the type of the fault; the integrated model includes multiple fault type identification models that have been trained, such as random forest algorithm, BP neural network, etc.

[0128] In the embodiment of the present application, the integrated model is used to comprehensively analyze the identification analysis results of multiple models, thereby improving the accuracy of model fault type identification.

[0129] Referring to Figure 3 The embodiment of the present application provides a residual current monitoring system for an AC system of a station, which comprises:

[0130] A signal receiving unit is configured to receive a current signal sent by a mobile terminal in a monitoring field, wherein the current signal comprises current signals of three phase lines and one neutral line.

[0131] A signal correction unit is configured to correct the received current signal based on a historical measurement result deviation change trend of a residual current sensing device by using a preset current signal correction model, wherein the current signal correction model comprises an encoder, a decoder and a mapping prediction model, the encoder is configured to extract current signal feature information, the decoder is configured to reconstruct the current signal based on the current signal feature information, and the mapping prediction model is configured to map the current signal feature information extracted by the encoder to current signal feature information after removing measurement deviation, and reconstruct the current signal after removing measurement deviation by using the decoder.

[0132] A signal synthesis unit is configured to perform vector synthesis based on the corrected current signals of the four lines to obtain a residual current signal.

[0133] A fault analysis unit is configured to extract a feature vector based on the current residual current signal, and analyze whether a fault occurs and a fault type based on the feature vector.

[0134] It should be noted that the residual current monitoring system for the AC system of the station provided in the embodiment is only used as an example for the division of the above functional units during residual current monitoring, and in actual applications, the above functions can be distributed to different functional units according to needs, that is, the internal structure of the device is divided into different functional units to complete all or part of the functions described above. In addition, the residual current monitoring system for the AC system of the station provided in the embodiment and the residual current monitoring method for the AC system of the station provided in the above embodiment belong to the same concept, and the specific implementation process is described in the method embodiment, which will not be repeated here.

[0135] The embodiment of the present application provides another residual current monitoring system for an AC system of a station, which comprises:

[0136] A residual current sensing device is in communication connection with a mobile terminal, configured to collect line current signals based on a signal collection instruction sent by the mobile terminal, and send the line current signals to the mobile terminal.

[0137] The mobile terminal is in communication connection with the cloud server and the residual current sensing device respectively, and is used for sending the current signal received from the residual current sensing device to the cloud server.

[0138] The cloud server is used for executing the residual current monitoring method of the station AC system in the foregoing embodiments to realize the residual current monitoring of the station AC system.

[0139] The embodiments of the present application provide a hierarchical deployment of the residual current sensing device, the mobile terminal and the cloud server, and joint implementation of the residual current monitoring of the station AC system and the fault analysis.

[0140] The embodiments of the present application provide a computer readable storage medium, which stores computer instructions, and the instructions are executed by a processor to realize the steps of the residual current monitoring method of the station AC system in the foregoing embodiments. It can be understood that the computer readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), a magnetic tape, a floppy disk and an optical data storage node, etc.

[0141] The present application is not limited to the above specific embodiments, and those skilled in the art can make various modifications without creative labor, which are all within the protection scope of the present application.

Claims

1. A method for monitoring residual current in a station AC system, characterized in that, include: The cloud server receives current signals sent by mobile terminals at the monitoring site. The current signals include current signals from three phase lines and one neutral line. Based on the historical measurement result deviation change trend of the residual current sensing device, a preset current signal correction model is used to correct the received current signal. The current signal correction model includes an encoder, a decoder, and a mapping prediction model. The encoder is used to extract current signal feature information, the decoder is used to reconstruct the current signal based on the current signal feature information, and the mapping prediction model is used to map the current signal feature information extracted by the encoder to the current signal feature information after removing the measurement deviation, and use the decoder to reconstruct the current signal after removing the measurement deviation from the current signal feature information. The remaining current signal is obtained by vector synthesis based on the current signals of the four corrected lines. Based on the current residual current signal, feature vectors are extracted, and the fault and its type are analyzed based on the feature vectors. The preset current signal correction model includes: a first encoder, used to input a current signal and acquire key features of the input signal; a first decoder, the first output of the first encoder being connected to the first input of the first decoder; used to generate a current signal based on the input key features; a first mapping prediction model, the second output of the first encoder being connected to the first input of the first mapping prediction model; used to acquire key features of the corresponding mapping based on the input key features; and a method for correcting the current signal using the current signal correction model, including: inputting the current signal currently collected by the current residual current sensing device into the first encoder, the first encoder outputting a third key feature; inputting the third key feature into the first mapping prediction model, the first mapping prediction model outputting a fourth key feature; and inputting the fourth key feature into the first decoder, the first decoder outputting a reconstructed current signal, i.e., the corrected current signal.

2. The method for monitoring residual current in a station AC system according to claim 1, characterized in that, The field mobile terminal is wirelessly or wiredly connected to the four residual current sensing devices. The communication method between the field mobile terminal and the residual current sensing devices includes: The on-site mobile terminal sends sampling synchronization signals and preset sampling durations to four residual current sensing devices; The residual current sensing device starts an analog-to-digital converter to collect the current signal input from the AC transformer based on the received sampling synchronization signal, and acquires the collected current signal after a preset sampling time. The residual current sensing device sends the collected current signal to the on-site mobile terminal.

3. The method for monitoring residual current in a station AC system according to claim 2, characterized in that, The field mobile terminal sends a sampling synchronization signal and a preset sampling duration to the residual current sensing device, including: The field mobile terminal sends an identical first reference signal to four residual current sensing devices; After receiving the first reference signal, the residual current sensing device generates the first reference signal reception completion time. The residual current sensing device sends the first reference signal reception completion time to the field mobile terminal to determine the reception delay error of the four residual current sensing devices; The on-site mobile terminal sends sampling synchronization signals, preset sampling duration, and reception delay duration to four residual current sensing devices.

4. The method for monitoring residual current in a station AC system according to claim 2, characterized in that, The residual current sensing device transmits the collected current signal to the on-site mobile terminal, including: Four residual current sensing devices splice a preset second reference signal with the collected current signal and send the spliced ​​signal to the on-site mobile terminal; The step of vector synthesis based on the corrected current signals of the four lines to obtain the residual current signal includes: aligning the current signals of the four lines, wherein the alignment includes: Based on the current signals of the four lines, a preset second reference signal is identified in the current signals of the four lines, and alignment is performed based on the preset second reference signal in the current signals of the four lines, thereby determining the alignment of the current signals collected by the four residual current sensing devices.

5. The method for monitoring residual current in a station AC system according to claim 1, characterized in that, The method for obtaining the preset current signal correction model includes: Based on the current signal historically collected by the residual current sensing device, it is input into a preset encoder network to obtain the encoded features. The encoded features are then input into a preset decoder network to obtain the decoded features, which are used to reconstruct the current signal. An error loss function is constructed based on the current signal input to the encoder network and the reconstructed current signal output by the decoder network. The encoder network and decoder network are iteratively trained based on the error loss function to obtain the first encoder and the first decoder after training. The measurement results based on the historical acquisition signals of the residual current sensing device are input into the first encoder to obtain the first key feature. The actual current signal based on the historical acquisition signals is input into the first encoder to obtain the second key feature. The first key feature is input into the preset mapping prediction model to obtain the output key feature. An error loss function is constructed based on the output key feature and the second key feature. The mapping prediction model is iteratively trained based on the error loss to obtain the trained first mapping prediction model.

6. The method for monitoring residual current in a station AC system according to claim 1, characterized in that, The process of inputting the fourth key feature into the first decoder and outputting the reconstructed current signal, i.e., the corrected current signal, includes: The current current signal collected by the current residual current sensing device is used as an additional reference signal for the reconstructed current signal output by the first decoder. The additional reference information and the fourth key feature are input into the first decoder, and the first decoder outputs the reconstructed current signal, which is the corrected current signal. The first decoder includes a self-attention unit and a cross-attention unit. An additional reference signal is input into the self-attention unit, and the key matrix K, query matrix Q, and value matrix V of the self-attention unit are calculated based on the additional reference signal to obtain the feature information of the additional reference signal. The feature information of the additional reference signal and the fourth key feature are jointly input into the cross-attention unit to obtain the decoded signal with the additional reference signal as a reference. The Q matrix of the cross-attention unit is calculated based on the fourth key feature as input, and the K and V matrices of the cross-attention unit are calculated based on the feature information of the additional reference signal as input.

7. The method for monitoring residual current in a station AC system according to claim 1, characterized in that, The process of extracting feature vectors based on the current residual current signal and analyzing whether a fault has occurred and the type of fault that has occurred based on the feature vectors includes: Based on the current residual current signal, extract time-domain features; obtain frequency distribution data by performing time-frequency transformation on the current residual current signal, and extract frequency-domain features based on the frequency distribution data; Based on time-domain and frequency-domain features, an integrated model is used to identify the types of faults that occur; the integrated model includes multiple fault type identification models that have already been trained.

8. A residual current monitoring system for a station AC system, characterized in that, include: The signal receiving unit is used to receive the current signal sent by the mobile terminal at the monitoring site. The current signal includes the current signals of the three phase lines and one neutral line. The signal correction unit is used to correct the received current signal based on the deviation change trend of the historical measurement results of the residual current sensing device using a preset current signal correction model. The current signal correction model includes an encoder, a decoder, and a mapping prediction model. The encoder is used to extract current signal feature information, the decoder is used to reconstruct the current signal based on the current signal feature information, and the mapping prediction model is used to map the current signal feature information extracted by the encoder to the current signal feature information after removing the measurement deviation, and use the decoder to reconstruct the current signal after removing the measurement deviation from the current signal feature information. The signal synthesis unit is used to perform vector synthesis based on the current signals of the four corrected lines to obtain the residual current signal; The fault analysis unit is used to extract feature vectors based on the current residual current signal, and to analyze whether a fault has occurred and the type of fault that has occurred based on the feature vectors. The signal correction unit includes a preset current signal correction model comprising: a first encoder for inputting a current signal and acquiring key features of the input signal; a first decoder, the first output of the first encoder being connected to the first input of the first decoder, for generating a current signal based on the input key features; and a first mapping prediction model, the second output of the first encoder being connected to the first input of the first mapping prediction model, for acquiring key features corresponding to the mapping based on the input key features. The method for correcting the current signal using the current signal correction model includes: inputting the current signal currently collected by the residual current sensing device into the first encoder, whereby the first encoder outputs a third key feature; inputting the third key feature into the first mapping prediction model, whereby the first mapping prediction model outputs a fourth key feature; and inputting the fourth key feature into the first decoder, whereby the first decoder outputs a reconstructed current signal, i.e., the corrected current signal.

9. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1-7.

Images