Ring main unit complete cycle operation data monitoring system based on artificial intelligence

Abstract

The invention relates to the technical field of ring main unit monitoring, and discloses a ring main unit full-cycle operation data monitoring system based on artificial intelligence, which comprises a topology self-discovery module used for constructing an adjacent matrix by calculating mutual impedance of physical units; the signal sequence construction module is used for constructing a signal sequence of each physical unit; the air pressure sequence construction module is used for constructing an air pressure sequence of each physical unit; the health scoring module is used for obtaining a health score of a physical unit corresponding to each node through a health scoring model; and the maintenance judgment module is used for calculating according to the health score of the physical unit to obtain an unrepaired probability. According to the method, the adjacency matrix is automatically constructed by injecting the current signals into the physical units in the ring main unit, and the health score of each physical unit is obtained by simultaneously performing correlation analysis on the signal sequences and the air pressure sequences of all the physical units through the health scoring model, so that the accuracy of evaluating the running state of the ring main unit is improved.

Description

Technical Field

[0001] The present invention relates to the field of ring main unit monitoring technology, and more specifically, to an artificial intelligence-based ring main unit full-cycle operation data monitoring system. Background Art

[0002] Ring main unit (RMU) is a type of high-voltage switchgear widely used in urban power distribution systems. It is primarily used for the access and distribution of electric energy. As a key node connecting substations and users in the power system, the health of the RMU is directly related to the operational safety and power supply reliability of the distribution network. Traditional RMU operation monitoring relies mainly on manual inspections, which is inefficient and difficult to detect potential hidden dangers in a timely manner.

[0003] With the rise of intelligent operations and maintenance (O&M), automated monitoring of ring main units (RMUs) is now achieved by collecting data from sensors such as voltage and current, constructing time series, and performing time series analysis. However, this single-data-source analysis method often overlooks the topological relationships between the various physical units within the RMU (such as circuit breakers and load switches). These units are electrically connected through a common busbar segment and exhibit certain electrical interdependencies. Therefore, relying solely on time series analysis from a single data source may fail to accurately capture the complex relationships between the physical units, thus affecting overall monitoring accuracy. Summary of the Invention

[0004] The present invention provides a ring main unit full-cycle operation data monitoring system based on artificial intelligence to solve the technical problems in the above-mentioned background technology.

[0005] The present invention provides a full-cycle operation data monitoring system for a ring main unit based on artificial intelligence, comprising: A topology self-discovery module, which is used to sequentially inject current signals into the physical units inside the ring main unit through small current injection modules integrated on the common busbar segment, and to construct an adjacency matrix by calculating the mutual impedance of the remaining physical units; The adjacency matrix is ​​a symmetric square matrix of size N×N, where N represents the number of nodes. Nodes correspond to physical units, and the element values ​​of the adjacency matrix are represented by 0 or 1. A signal sequence building module is used to collect the instantaneous waveform signals of the voltage and current of each physical unit, and calculate the signal vector as a sequence unit of the signal sequence based on the complex vector at each moment; The signal vector consists of voltage amplitude, voltage phase angle, current amplitude, current phase angle and voltage-current phase difference; A pressure sequence building module is used to collect the sulfur hexafluoride pressure of each physical unit and correct the pressure at each moment through thermal-pressure coupling compensation to obtain a pressure correction value as a sequence unit of the pressure sequence; A health scoring module is used to input the adjacency matrix, the signal sequence of the node, and the air pressure sequence into the health scoring model, and output the health score of the physical unit corresponding to each node; The maintenance judgment module is used to calculate the disrepair probability based on the health score of the physical unit. If the disrepair probability is greater than or equal to a preset probability threshold, it means that the physical unit needs maintenance.

[0006] Furthermore, the current signal injected into the physical unit is a 100 Hz pseudo-random binary sequence of 0.5% of the rated current of the ring main unit. The voltage signals of the other physical units are synchronously monitored, and the voltage response amplitude at the excitation frequency is extracted. The mutual impedance between any two physical units is equal to the ratio of the voltage response amplitude to the injected current amplitude. If the absolute value of the mutual impedance is less than the critical mutual impedance threshold, the element value between the corresponding two nodes in the adjacency matrix is ​​1, otherwise it is 0. The critical mutual impedance threshold is a custom parameter.

[0007] Furthermore, the complex coefficients of the 50 Hz components of the instantaneous waveform signals of the voltage and current are respectively extracted by Fourier transform within a preset time window as a complex vector at a moment, wherein the size of the preset time window is a custom parameter, the complex vector is composed of a first voltage complex coefficient, a second voltage complex coefficient, a first current complex coefficient, and a second current complex coefficient, the calculation formulas for the voltage amplitude and the current amplitude are the same, the calculation formulas for the voltage phase angle and the current phase angle are the same, and the voltage-current phase difference is equal to the difference between the voltage phase angle and the current phase angle; Voltage amplitude The calculation formula is as follows: ,in represents the first voltage complex coefficient, represents the second voltage complex coefficient; Voltage phase angle The calculation formula is as follows: .

[0008] Furthermore, the calculation formula for thermal-pressure coupling compensation is as follows: ,in Indicates the air pressure correction value, Indicates the original air pressure before correction. represents the reference temperature, Indicates the measured temperature. represents the zero drift compensation term, represents the thermal drift coefficient, represents the thermal compensation balance factor, Represents the gas residual correction term.

[0009] Furthermore, the health scoring model consists of a first sequence analysis layer, a second sequence analysis layer, a feature fusion layer, a topology analysis layer, and a classifier; The first sequence analysis layer inputs the signal sequence of each node and outputs the first vector; The second sequence analysis layer inputs the air pressure sequence of each node and outputs the second vector; The feature fusion layer is used to fuse the first vector and the second vector of the node to obtain the third vector; The topology analysis layer is used to update the third vector of the node according to the adjacency matrix to obtain the fourth vector; The fourth vector of each node is input into the classifier, and the category space of the classifier represents the health score of the physical unit corresponding to each node; The dimensions of the first vector, the second vector, the third vector, and the fourth vector are all custom parameters.

[0010] Furthermore, the calculation steps of the first sequence analysis layer include: Step S201, performing wavelet decomposition on the transposed signal sequence by discrete wavelet transform to obtain a first characteristic sequence; The length of the first feature sequence is , each sequence unit is represented by a vector with a dimension of 5, where 1≤i≤L, A represents the length of the signal sequence, and L represents the number of decomposition layers of the discrete wavelet transform. Indicates rounding down, the range of the number of decomposition layers of discrete wavelet transform is [1, ]; Step S202: performing time series analysis on the first feature sequence using a Transformer model to obtain a second feature sequence; The length of the second feature sequence is the same as that of the first feature sequence, and the number of dimensions of each sequence unit is a custom parameter; Step S203, performing pooling processing on the second feature sequence by average pooling to obtain a feature vector, and converting the feature vector into a first vector by nonlinear mapping; The calculation formula of nonlinear mapping is as follows: ,in Represents the first vector, X represents the eigenvector, W represents the weight matrix, b represents the bias vector, and Softplus represents the Softplus activation function.

[0011] Furthermore, the second sequence analysis layer is built based on the LSTM model; The calculation formula of the feature fusion layer is as follows: ,in 、 and represent the first, second and third vectors respectively, and represent the first weight matrix and the first bias vector respectively, and Represent the second weight matrix and the second bias vector respectively, tanh represents the hyperbolic tangent activation function, Represents the Sigmoid activation function, and ⊙ represents the Hadamard product.

[0012] Furthermore, the topology analysis layer is constructed based on the GAT model.

[0013] Furthermore, the calculation formula of disrepair probability is as follows: , where 1≤i≤N, represents the health score of the physical unit corresponding to the i-th node, e represents a natural constant, and the preset probability threshold is a custom parameter.

[0014] Furthermore, the input and output of the health scoring model corresponding to the disrepair probability false alarm are recorded to construct a feedback data set, and the relative entropy is regularly judged to be greater than the preset relative entropy threshold. Then, the health scoring model is incrementally trained based on the feedback data set, where the preset relative entropy threshold is a custom parameter; The calculation formula of relative entropy S is as follows: ,in represents the probability distribution corresponding to the nth physical unit in the feedback data set, Represents the predicted probability distribution of the nth physical unit in the health score model.

[0015] The beneficial effects of the present invention are as follows: the present invention automatically constructs an adjacency matrix by injecting current signals into the physical units inside the ring network cabinet, and simultaneously performs correlation analysis on the signal sequences and air pressure sequences of all physical units through the health scoring model to obtain the health score of each physical unit, thereby improving the accuracy of the evaluation of the operating status of the ring network cabinet. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of a full-cycle operation data monitoring system for a ring main unit based on artificial intelligence according to the present invention; Figure 2 It is a calculation flow chart of the first sequence analysis layer of the present invention.

[0017] In the figure: topology self-discovery module 101, signal sequence construction module 102, air pressure sequence construction module 103, health scoring module 104, and maintenance judgment module 105. DETAILED DESCRIPTION

[0018] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0019] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprising" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0020] like Figures 1 and 2 As shown, an artificial intelligence-based ring main unit full-cycle operation data monitoring system includes: A topology self-discovery module 101 is used to sequentially inject current signals into the physical units inside the ring main unit through small current injection modules integrated on the common busbar segment, and to construct an adjacency matrix by calculating the mutual impedance of the remaining physical units; The adjacency matrix is ​​a symmetric square matrix of size N×N, where N represents the number of nodes. Nodes correspond to physical units, and the element values ​​of the adjacency matrix are represented by 0 or 1. A signal sequence building module 102 is used to collect the instantaneous waveform signals of the voltage and current of each physical unit, and calculate the signal vector as a sequence unit of the signal sequence based on the complex vector at each moment; The signal vector consists of voltage amplitude, voltage phase angle, current amplitude, current phase angle and voltage-current phase difference; The air pressure sequence construction module 103 is used to collect the sulfur hexafluoride air pressure of each physical unit and correct the air pressure at each moment through thermal pressure coupling compensation to obtain an air pressure correction value as a sequence unit of the air pressure sequence; A health scoring module 104 is configured to input the adjacency matrix, the signal sequence of the node, and the pressure sequence into a health scoring model, and output a health score of the physical unit corresponding to each node; The maintenance judgment module 105 is used to calculate the disrepair probability according to the health score of the physical unit, and if the disrepair probability is greater than or equal to a preset probability threshold, it indicates that the physical unit needs maintenance.

[0021] In one embodiment of the present invention, the current signal injected into the physical unit is a 100 Hz pseudo-random binary sequence (PRBS) of 0.5% of the rated current of the ring main unit. The other physical units synchronously monitor the voltage signal, and the voltage response amplitude at the excitation frequency is extracted. The mutual impedance between any two physical units is equal to the ratio of the voltage response amplitude to the injected current amplitude. If the absolute value of the mutual impedance is less than the critical mutual impedance threshold, the element value between the corresponding two nodes in the adjacency matrix is ​​1 (indicating that there is an electrical connection relationship between the two physical units), otherwise it is 0 (indicating that there is no electrical connection relationship between the two physical units). The critical mutual impedance threshold is a custom parameter. For example, if the rated current of the ring main unit is 400 A, the current value of the injected current signal is 2 A, and the critical mutual impedance threshold can be set to 0.02 Ω. The physical unit can be a load switch, a vacuum circuit breaker, etc.

[0022] Mutual impedance between the i-th physical unit and the j-th physical unit The calculation formula is as follows: , where 1≤i≤N, 1≤j≤N, represents the voltage response amplitude of the jth physical unit, Represents the injection current amplitude of the i-th physical unit.

[0023] In one embodiment of the present invention, the frequency of collecting the instantaneous waveform signals of voltage and current and the length of the signal sequence are both custom parameters. Preferably, the collection frequency is set to 10 kHz and the length of the signal sequence is set to 50.

[0024] In one embodiment of the present invention, the complex coefficients of the 50 Hz component of the instantaneous waveform signals of the voltage and current are respectively extracted by Fourier transform within a preset time window as a complex vector at a moment, wherein the size of the preset time window is a custom parameter. Preferably, the preset time window size is set to 20 ms, that is, the number of Fourier transform points is 200, and the complex vector is composed of a first voltage complex coefficient, a second voltage complex coefficient, a first current complex coefficient, and a second current complex coefficient. The calculation formulas for the voltage amplitude and the current amplitude are the same, the calculation formulas for the voltage phase angle and the current phase angle are the same, and the voltage-current phase difference is equal to the difference between the voltage phase angle and the current phase angle.

[0025] Voltage amplitude The calculation formula is as follows: ,in represents the first voltage complex coefficient, represents the second voltage complex coefficient; Voltage phase angle The calculation formula is as follows: .

[0026] For example, if the first voltage complex coefficient is 162.6 and the second voltage complex coefficient is 9.1, then the voltage amplitude calculated according to the above calculation formula is approximately 162.85 V and the voltage phase angle is approximately -3.2°.

[0027] It should be noted that in the power system, the operating basis of all electrical equipment is the industrial frequency power supply, whose frequency is 50Hz. The purpose of extracting the complex coefficients of the 50Hz component of the instantaneous waveform signals of voltage and current is to obtain the steady-state operation information of the ring network cabinet and remove the influence of transient noise, high-frequency disturbances and harmonics.

[0028] In one embodiment of the present invention, the length of the pressure sequence is a custom parameter. Preferably, the length of the pressure sequence is set to 12. For example, the pressure of sulfur hexafluoride is collected at intervals of 5 minutes within an hour or at intervals of 2 hours within a day. In addition, the present invention uses sulfur hexafluoride (SF6) as an insulating and arc-extinguishing medium. Under the action of an electric arc and high temperature conditions, sulfur hexafluoride can decompose into toxic substances. If leakage occurs, it will pose a threat to the environment and human health, and affect the insulation performance of the physical units inside the ring network cabinet, which may cause a short circuit or electrical failure. Therefore, sulfur hexafluoride pressure monitoring is particularly important.

[0029] In one embodiment of the present invention, the calculation formula for thermal pressure coupling compensation is as follows: ,in Indicates the air pressure correction value, Indicates the original air pressure before correction. represents the reference temperature, Indicates the measured temperature. represents the zero drift compensation term, represents the thermal drift coefficient, represents the thermal compensation balance factor, Represents the gas residual correction term.

[0030] It should be noted that the reference temperature is a custom parameter. Preferably, the reference temperature is set to 20°C (converted to 293.15K); the zero drift compensation item represents the output drift of the correction pressure sensor in a no-pressure / low-pressure state, which can be obtained by sensor calibration or manufacturer's standardization, and the unit is kPa; the thermal drift coefficient represents the compensation for the stress change of the pressure sensor caused by the temperature increase, and the unit is kPa / K. For example, if the thermal drift coefficient is set to 0.2kPa / K, it means that for every 1K increase in temperature, the pressure sensor outputs 0.2kPa more; the thermal compensation balance factor is a custom parameter to prevent the denominator from being 0 or close to 0, and the unit is K. Preferably, the thermal compensation balance factor is set to 20K; the gas residual correction item is a custom parameter, which represents a constant offset item used to eliminate the calibration error, and the unit is kPa. Preferably, the gas residual correction item is set to 1.5kPa.

[0031] For example, if the zero drift compensation term is 1.5 kPa, the original air pressure before correction is 320 kPa, and the measured temperature is 308.15 K (35°C), then according to the above calculation formula, the air pressure correction value can be obtained to be approximately 302.58 kPa.

[0032] In one embodiment of the present invention, the health scoring model consists of a first sequence analysis layer, a second sequence analysis layer, a feature fusion layer, a topology analysis layer, and a classifier; The first sequence analysis layer inputs the signal sequence of each node and outputs the first vector; The second sequence analysis layer inputs the air pressure sequence of each node and outputs the second vector; The feature fusion layer is used to fuse the first vector and the second vector of the node to obtain the third vector; The topology analysis layer is used to update the third vector of the node according to the adjacency matrix to obtain the fourth vector; The fourth vector of each node is input into the classifier, and the category space of the classifier represents the health score of the physical unit corresponding to each node; The number of dimensions of the first vector, the second vector, the third vector and the fourth vector are all custom parameters. Preferably, the number of dimensions of the first vector, the second vector, the third vector and the fourth vector are set to 64, 32, 128 and 256 respectively.

[0033] In one embodiment of the present invention, the value range of the health score is [1, 10], and a larger value of the health score indicates a better health of the physical unit, and the sample labels (i.e., health scores) of the training samples used to train the health score model can be obtained through evaluation by relevant experts.

[0034] In one embodiment of the present invention, Figure 2 As shown, the calculation steps of the first sequence analysis layer include: Step S201, performing wavelet decomposition on the transposed signal sequence by discrete wavelet transform to obtain a first characteristic sequence; The jth column of the transposed signal sequence is expressed as follows: , where 1≤j≤5, Represent the j-th dimension value of the 1st to A-1th and A-th sequence units of the signal sequence respectively; The result of applying the L-layer discrete wavelet transform is expressed as follows: ,in represents the approximate coefficient of the Lth layer for the jth dimension value, Represent the detail coefficients of the Lth layer, the L-1th layer to the 1st layer of the jth dimension value respectively; The length of the approximation coefficient of the Lth layer is , where A represents the length of the signal sequence, L represents the number of decomposition layers of discrete wavelet transform, Indicates rounding down; The length of the detail coefficient of layer i is , where 1≤i≤L; The range of the number of decomposition layers of discrete wavelet transform is [1, ], preferably, the number of decomposition layers of discrete wavelet transform is set to 2. For example, if the length of the signal sequence is set to 50, the length of the approximate coefficient of the second layer is 12, the length of the detail coefficient of the second layer is 12, and the length of the detail coefficient of the first layer is 25. The total amount of data is (12+12+25)×5=245; That is, the length of the first feature sequence is , each sequence unit is represented by a vector with a dimension of 5; Step S202: performing time series analysis on the first feature sequence using a Transformer model to obtain a second feature sequence; The second feature sequence has the same length as the first feature sequence, and the number of dimensions of each sequence unit is a custom parameter. Preferably, the number of dimensions of each sequence unit of the second feature sequence is set to 16; Step S203, performing pooling processing on the second feature sequence by average pooling to obtain a feature vector, and converting the feature vector into a first vector by nonlinear mapping; The calculation formula of nonlinear mapping is as follows: ,in represents the first vector, X represents the eigenvector, W represents the weight matrix (16×64), b represents the bias vector (1×64), and Softplus represents the Softplus activation function.

[0035] In one embodiment of the present invention, the second sequence analysis layer is constructed based on an LSTM (long short-term memory recurrent network) model. Specifically, the number of hidden layers in the second sequence analysis layer is the same as the length of the air pressure sequence, the input of each hidden layer corresponds one-to-one to the sequence unit of the air pressure sequence, and the output of the last hidden layer is used as the second vector. The second sequence analysis layer can also be constructed based on a GRU (gated recurrent network) model, which will not be elaborated here.

[0036] The calculation formula of the feature fusion layer is as follows: ,in 、 and represent the first, second and third vectors respectively, and Represent the first weight matrix (64×128) and the first bias vector (1×128), and Represent the second weight matrix (32×128) and the second bias vector (1×128), tanh represents the hyperbolic tangent activation function, Represents the Sigmoid activation function, and ⊙ represents the Hadamard product.

[0037] In one embodiment of the present invention, the topology analysis layer is constructed based on the GAT (graph attention network) model, and can also be constructed based on the GNN (graph neural network) model, which will not be described in detail here.

[0038] In one embodiment of the present invention, the calculation formula for the disrepair probability is as follows: , where 1≤i≤N, represents the health score of the physical unit corresponding to the i-th node, e represents a natural constant, and the preset probability threshold is a custom parameter. Preferably, the preset probability threshold is set to 0.4.

[0039] In one embodiment of the present invention, when the operation and maintenance personnel conduct offline maintenance, if they find that the probability of disrepair is falsely reported, they will record the input and output of the health scoring model to construct a feedback data set, and regularly calculate the relative entropy. If the relative entropy is greater than a preset relative entropy threshold, the health scoring model will be incrementally trained based on the feedback data set, where the preset relative entropy threshold is a custom parameter. Preferably, the preset relative entropy threshold is set to 0.05; The calculation formula of relative entropy S is as follows: ,in represents the probability distribution corresponding to the nth physical unit in the feedback data set, Represents the predicted probability distribution of the nth physical unit in the health score model.

[0040] For example, the ring main unit includes three physical units. The probability distribution is represented by the normalized health score. The predicted probability distributions of the three physical units in the health score model are 0.7, 0.2, and 0.1, respectively. The corresponding probability distributions of the three physical units in the feed data set are 0.5, 0.3, and 0.2, respectively. According to the above calculation formula, the relative entropy is approximately 0.09.

[0041] It should be noted that the incremental training of the health scoring model enables the model to continuously adapt to the dynamically changing data distribution in real scenarios, giving the model the ability to "self-recover" and "continuously learn" to prevent performance degradation or overfitting of old data.

[0042] It should be noted that the intervals and thresholds are set for ease of comparison. The threshold size depends on the amount of sample data and the cardinality set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless numerical calculations. These formulas are derived from software simulations of the most recent real-world conditions using large amounts of data. The preset parameters in these formulas are set by those skilled in the art based on actual conditions.

[0043] The above describes the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation methods. The above specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. An artificial intelligence-based ring main unit full-cycle operation data monitoring system, characterized in that: include: A topology self-discovery module, which is used to sequentially inject current signals into the physical units inside the ring main unit through small current injection modules integrated on the common busbar segment, and to construct an adjacency matrix by calculating the mutual impedance of the remaining physical units; The adjacency matrix is ​​a symmetric square matrix of size N×N, where N represents the number of nodes. Nodes correspond to physical units, and the element values ​​of the adjacency matrix are represented by 0 or 1. A signal sequence building module is used to collect the instantaneous waveform signals of the voltage and current of each physical unit, and calculate the signal vector as a sequence unit of the signal sequence based on the complex vector at each moment; The signal vector consists of voltage amplitude, voltage phase angle, current amplitude, current phase angle and voltage-current phase difference; A pressure sequence building module is used to collect the sulfur hexafluoride pressure of each physical unit and correct the pressure at each moment through thermal-pressure coupling compensation to obtain a pressure correction value as a sequence unit of the pressure sequence; A health scoring module is used to input the adjacency matrix, the signal sequence of the node, and the air pressure sequence into the health scoring model, and output the health score of the physical unit corresponding to each node; The maintenance judgment module is used to calculate the disrepair probability based on the health score of the physical unit. If the disrepair probability is greater than or equal to a preset probability threshold, it means that the physical unit needs maintenance.

2. The artificial intelligence-based ring main unit full-cycle operation data monitoring system according to claim 1 is characterized in that: The current signal injected into the physical unit is a 100 Hz pseudo-random binary sequence of 0.5% of the rated current of the ring main unit. The other physical units synchronously monitor the voltage signals and extract the voltage response amplitude at the excitation frequency. The mutual impedance between any two physical units is equal to the ratio of the voltage response amplitude to the injected current amplitude. If the absolute value of the mutual impedance is less than the critical mutual impedance threshold, the element value between the corresponding two nodes in the adjacency matrix is ​​1; otherwise, it is 0. The critical mutual impedance threshold is a custom parameter.

3. The artificial intelligence-based ring main unit full-cycle operation data monitoring system according to claim 1 is characterized in that: Extract the complex coefficients of the 50 Hz components of the instantaneous waveform signals of the voltage and current respectively through Fourier transform within a preset time window as a complex vector at a moment, wherein the size of the preset time window is a custom parameter, and the complex vector consists of a first voltage complex coefficient, a second voltage complex coefficient, a first current complex coefficient, and a second current complex coefficient. The calculation formulas for the voltage amplitude and the current amplitude are the same, the calculation formulas for the voltage phase angle and the current phase angle are the same, and the voltage-current phase difference is equal to the difference between the voltage phase angle and the current phase angle; Voltage amplitude The calculation formula is as follows: ,in represents the first voltage complex coefficient, represents the second voltage complex coefficient; Voltage phase angle The calculation formula is as follows: .

4. The artificial intelligence-based ring main unit full-cycle operation data monitoring system according to claim 1 is characterized in that: The calculation formula for thermal pressure coupling compensation is as follows: ,in Indicates the air pressure correction value, Indicates the original air pressure before correction. represents the reference temperature, Indicates the measured temperature. represents the zero drift compensation term, represents the thermal drift coefficient, represents the thermal compensation balance factor, Represents the gas residual correction term.

5. The artificial intelligence-based ring main unit full-cycle operation data monitoring system according to claim 1 is characterized in that: The health scoring model consists of the first sequence analysis layer, the second sequence analysis layer, the feature fusion layer, the topology analysis layer and the classifier; The first sequence analysis layer inputs the signal sequence of each node and outputs the first vector; The second sequence analysis layer inputs the air pressure sequence of each node and outputs the second vector; The feature fusion layer is used to fuse the first vector and the second vector of the node to obtain the third vector; The topology analysis layer is used to update the third vector of the node according to the adjacency matrix to obtain the fourth vector; The fourth vector of each node is input into the classifier, and the category space of the classifier represents the health score of the physical unit corresponding to each node; The dimensions of the first vector, the second vector, the third vector, and the fourth vector are all custom parameters.

6. The artificial intelligence-based ring main unit full-cycle operation data monitoring system according to claim 5 is characterized in that: The computational steps of the first sequence analysis layer include: Step S201, performing wavelet decomposition on the transposed signal sequence by discrete wavelet transform to obtain a first characteristic sequence; The length of the first feature sequence is , each sequence unit is represented by a vector with a dimension of 5, where 1≤i≤L, A represents the length of the signal sequence, and L represents the number of decomposition layers of the discrete wavelet transform. Indicates rounding down, the range of the number of decomposition layers of discrete wavelet transform is [1, ]; Step S202: performing time series analysis on the first feature sequence using a Transformer model to obtain a second feature sequence; The length of the second feature sequence is the same as that of the first feature sequence, and the number of dimensions of each sequence unit is a custom parameter; Step S203, performing pooling processing on the second feature sequence by average pooling to obtain a feature vector, and converting the feature vector into a first vector by nonlinear mapping; The calculation formula of nonlinear mapping is as follows: ,in Represents the first vector, X represents the eigenvector, W represents the weight matrix, b represents the bias vector, and Softplus represents the Softplus activation function.

7. The artificial intelligence-based ring main unit full-cycle operation data monitoring system according to claim 5 is characterized in that: The second sequence analysis layer is built based on the LSTM model; The calculation formula of the feature fusion layer is as follows: ,in 、 and represent the first, second and third vectors respectively, and represent the first weight matrix and the first bias vector respectively, and Represent the second weight matrix and the second bias vector respectively, tanh represents the hyperbolic tangent activation function, Represents the Sigmoid activation function, and ⊙ represents the Hadamard product.

8. The artificial intelligence-based ring main unit full-cycle operation data monitoring system according to claim 5 is characterized in that: The topology analysis layer is built based on the GAT model.

9. The artificial intelligence-based ring main unit full-cycle operation data monitoring system according to claim 1 is characterized in that: The formula for calculating the probability of disrepair is as follows: , where 1≤i≤N, represents the health score of the physical unit corresponding to the i-th node, e represents a natural constant, and the preset probability threshold is a custom parameter.

10. The artificial intelligence-based ring main unit full-cycle operation data monitoring system according to claim 1 is characterized in that: Record the input and output of the health scoring model corresponding to the disrepair probability false alarm to build a feedback data set, and regularly determine if the relative entropy is greater than the preset relative entropy threshold. Then, incremental training of the health scoring model is performed based on the feedback data set, where the preset relative entropy threshold is a custom parameter; The calculation formula of relative entropy S is as follows: ,in represents the probability distribution corresponding to the nth physical unit in the feedback data set, Represents the predicted probability distribution of the nth physical unit in the health score model.

Images