Intelligent monitoring method and system for running state of low-voltage transformer area

Abstract

The invention discloses an intelligent monitoring method and system for the running state of a low-voltage transformer area, and relates to the technical field of power system monitoring, and the method comprises the following steps: S1, collecting multi-source heterogeneous running data through sensing equipment disposed in the low-voltage transformer area; s2, performing data cleaning and time alignment on the multi-source heterogeneous operation data, and performing normalization to obtain regular time sequence data; s3, on the basis of the regular time sequence data, constructing a transformer area health degree comprehensive calculation model, and calculating to obtain a health degree index of each transformer area; s4, generating graded early warning information according to the numerical range of the health degree index of the transformer area; s5, predicting the low voltage trend of each node of the transformer area in the future time period based on the historical voltage data and the transformer area topological structure information; and S6, carrying out storage and visual display on the collected and predicted data. According to the invention, through data analysis and prediction, deep perception and fine management of the operation state of the transformer area are realized.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of power system monitoring, in particular to a low-voltage transformer area operation state intelligent monitoring method and system. BACKGROUND

[0002] A transformer area refers to an area powered by a power distribution transformer in a power system, and is a term in the economic operation and management of power, which is also used as a place name in a specific context. The physical range covers the line, equipment and covered area between the transformer low-voltage side outlet and the user electric energy meter, and is the basic unit for power supply enterprises to carry out power distribution management, line loss accounting and user service.

[0003] The operation state of the low-voltage power distribution transformer area directly determines the power quality and power supply reliability of the user as the last stage of power transmission of the power grid. With the development of social economy, the user's requirements for power quality and power supply continuity are increasing, so it is crucial to monitor the low-voltage transformer area in real time, accurately and intelligently. Effective monitoring can timely find hidden dangers such as low voltage, equipment overload and three-phase imbalance, and is the basis for realizing rapid response to faults, improving operation and maintenance efficiency and ensuring safe and stable operation of the power grid.

[0004] A low-voltage transformer area state monitoring method is disclosed in Chinese Patent No. CN116566035A, which comprises S1 mobile terminal detection to obtain network structure data of the low-voltage transformer area and upload to the master station; S2 master station to call low-voltage measurement data and combine the network structure data of the low-voltage transformer area to establish a low-voltage power grid quality analysis model; S3 to monitor the state of the low-voltage transformer area by using the low-voltage power grid quality analysis model and solve abnormal state. The application aims to solve the problem of low-voltage transformer area power grid quality, and establishes a low-voltage power grid analysis model by combining the network structure of the low-voltage transformer area with the measurement data, analyzes the low-voltage weak points by using the model, guides the optimization of the low-voltage network structure, solves the problems of low voltage, high loss and frequent faults, and provides support for subsequent transformer area load, photovoltaic access and project planning.

[0005] A low-voltage transformer area operation state intelligent monitoring system is disclosed in Chinese Patent No. CN111026927B, which comprises a data acquisition module for acquiring operation parameter data; a topology identification data analysis module for statistical analysis by using a topology identification calculation model, identifying or verifying box meter relationship and household transformer relationship, and realizing topology identification mapping; and a topology identification imaging module for visual display of data analysis results, dynamically generating a topology map and real-time tracking and verification, supporting dynamic and conditional query of transformer area management, transformer area topology, abnormality and analysis log. The application can effectively realize the power grid safety power utilization management and safety operation and maintenance monitoring mode, and provide reliable basis for overall planning and overall management of the power grid.

[0006] The above patents all have the problems proposed in the background: most of the evaluation methods are single index threshold judgment or simple static weighted scoring, which cannot comprehensively and dynamically reflect the overall health level of the transformer area; and most of them are post-alarm, which cannot realize prospective early warning of faults. SUMMARY

[0007] The technical problem to be solved by the present application is to provide a low-voltage transformer area operation state intelligent monitoring method and system to solve the problems of the prior art.

[0008] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is:

[0009] A low-voltage transformer area operation state intelligent monitoring method, comprising the following steps:

[0010] Step S1, collecting multi-source heterogeneous operation data through a sensing device deployed in a low-voltage transformer area, wherein the multi-source heterogeneous operation data includes electrical data, event data, state data and environmental data;

[0011] Step S2, data cleaning and time alignment are performed on the multi-source heterogeneous operation data, and normalized to obtain regular time series data;

[0012] Step S3, based on the regular time series data, a transformer area health degree comprehensive calculation model is constructed, and the health degree index of each transformer area is calculated;

[0013] Step S4, generating graded early warning information according to the numerical range of the transformer area health degree index;

[0014] Step S5, based on historical voltage data and transformer area topology structure information, predicting the low-voltage trend of each node of the transformer area in the future period;

[0015] Step S6, storing and visualizing the early warning information, multi-source heterogeneous operation data, health degree index and low-voltage trend prediction results.

[0016] Further, in step S1, the electrical data includes three-phase voltage, three-phase current and zero-sequence current; the event data includes switch tripping signal and protection action event; the state data includes terminal online state; and the environmental data includes transformer oil temperature.

[0017] Further, in step S2, the data cleaning specifically includes: using an adaptive weighted sliding window filtering algorithm, wherein each data point in the sliding window is given an adaptive weight, which is determined by a time decay factor and the mutation degree of the point data relative to the neighborhood data, and the final cleaned data value is the weighted average value of all data in the window.

[0018] Further, in the step S2, the time alignment specifically includes: using a dynamic time warping algorithm to perform time alignment on the multi-source data.

[0019] Further, in the step S3, the calculation manner of the transformer area health degree comprehensive calculation model is: a health degree basic index and a penalty factor are respectively constructed, and the health degree basic index and the penalty factor of each transformer area are multiplied to obtain the final health degree index of each transformer area.

[0020] The health degree basic index includes: a voltage quality index term, a communication state index term, a device event index term, and a load balancing index term.

[0021] The penalty factor includes: a temperature-related penalty term and a voltage-related penalty term, wherein the temperature-related penalty term is composed of a temperature penalty coefficient and a transformer oil temperature out-of-limit indication function, and the voltage-related penalty term is composed of a voltage penalty coefficient and a voltage quality indication function.

[0022] Further, in the step S4, the specific process of generating the hierarchical early warning information is: four health degree index intervals are pre-set, when the health degree index is greater than or equal to 0.8 and less than or equal to 1.0, it corresponds to a first-level normal early warning, when the health degree index is greater than or equal to 0.6 and less than 0.8, it corresponds to a second-level attention early warning, when the health degree index is greater than or equal to 0.4 and less than 0.6, it corresponds to a third-level abnormal early warning, and when the health degree index is greater than or equal to 0 and less than 0.4, it corresponds to a fourth-level serious early warning; the system calculates the health degree index in real time and judges the interval to which it belongs, and automatically generates the early warning signal of the corresponding level.

[0023] Further, in the step S5, the prediction of the low-voltage trend of each node of the transformer area in the future period specifically includes: a directed graph structure constructed according to the physical topology connection relationship of the transformer area, and historical voltage time series data corresponding to each node in the graph as model input; a spatio-temporal graph convolution network model is used for prediction, the spatio-temporal graph convolution network model includes 3 cascaded spatio-temporal convolution blocks; each spatio-temporal convolution block is composed of a graph convolution layer in the spatial dimension and a one-dimensional convolution layer in the time dimension connected in series; after the 3 spatio-temporal convolution blocks are stacked, the final voltage prediction value sequence corresponding to each node in a specific period in the future is obtained through a fully connected output layer.

[0024] Further, in the spatio-temporal convolution block, the graph convolution layer adopts a spectral graph convolution method based on Chebyshev polynomial approximation, and the order K of the Chebyshev graph polynomial is set to 3; the one-dimensional convolution layer adopts a dilated causal convolution structure, and the convolution kernel size of the dilated causal convolution layer is set to 3, and the dilated coefficients thereof are respectively set to 1, 2, and 4 in the 3 blocks.

[0025] A low-voltage transformer area operation state intelligent monitoring system, comprising:

[0026] A data acquisition module is configured to acquire multi-source heterogeneous operation data from various intelligent terminal devices in a low-voltage transformer area.

[0027] A data preprocessing module is configured to perform an adaptive weighted sliding window filtering algorithm to clean and align the data.

[0028] A health degree calculation module is configured to calculate a dynamically weighted transformer area health degree index.

[0029] An intelligent early warning module is configured to generate graded early warning information according to the health degree index.

[0030] A space-time prediction module is configured to perform low-voltage trend prediction based on a space-time graph convolution network model.

[0031] A data storage and visualization module is configured to store all data and analysis results and provide a graphical human-computer interaction interface.

[0032] Further, the health degree calculation module comprises:

[0033] An entropy weight method calculation unit is configured to dynamically calculate weight coefficients of voltage quality, communication status, device event and load balancing indexes according to historical data volatility of each index.

[0034] A nonlinear transformation unit is configured to perform nonlinear mapping processing on the voltage quality index and the device event index by using a Sigmoid function.

[0035] A penalty factor calculation unit is configured to calculate a multiplication penalty factor according to transformer oil temperature overrun and voltage quality index overrun.

[0036] Compared with the prior art, the present application has the following advantages:

[0037] 1. The present application overcomes the contradiction between effectively removing noise and retaining real mutation characteristics of data by using the adaptive weighted sliding window filtering algorithm. The algorithm can automatically identify and smooth random noise while sensitively retaining real fault characteristics such as voltage sag and current surge, providing a high-quality and reliable data basis for subsequent advanced analysis and greatly improving the accuracy of the monitoring system.

[0038] 2. The dynamic weighting health degree index model proposed in the present application innovatively integrates multi-dimensional indexes such as voltage quality, communication status, device event and load balancing, and introduces an entropy weight method to dynamically adjust the weight, so that the evaluation result can objectively reflect the actual importance change of each index. In addition, the introduction of nonlinear transformation and penalty factor makes the evaluation result not only comprehensive but also more sensitive to serious abnormalities.

[0039] 3、The application uses a space-time graph convolution network to predict low-voltage trends, breaking through the limitation of traditional time series prediction methods that only consider the time dimension and ignore the spatial topological relationship. The model can simultaneously capture the evolution law of the voltage of each monitoring point in the transformer area over time and the spatial correlation and propagation influence on the electrical topology, thereby realizing earlier, more accurate, and more spatially identifiable prediction of low-voltage risks, providing key technical support for deploying voltage regulation measures in advance and avoiding low-voltage occurrence.

[0040] 4、The application integrates data acquisition, cleaning, evaluation, early warning, prediction and other functional modules into a whole system that works cooperatively. The system not only provides full-cycle insight from real-time monitoring to long-term prediction, but also provides centralized display through a visual interface, greatly improving the overall perception ability of the operation and maintenance personnel to the low-voltage transformer area state and the operation and maintenance efficiency, realizing intelligent and lean operation and maintenance of the low-voltage distribution network. BRIEF DESCRIPTION OF DRAWINGS

[0041] Other features, objects and advantages of the application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings:

[0042] Figure 1 A flowchart of an embodiment of the application;

[0043] Figure 2 A system module diagram of an embodiment of the application;

[0044] Figure 3 A space-time graph convolution network diagram of an embodiment of the application;

[0045] Figure 4 A visual system diagram of an embodiment of the application. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical scheme and advantages of the application more clear, the application will be described in detail below with reference to the drawings and specific embodiments.

[0047] As shown in Figure 1 A low-voltage transformer area operation state intelligent monitoring method, comprising the following steps:

[0048] Step S1, collecting multi-source heterogeneous operation data through a sensing device deployed in the low-voltage transformer area, wherein the multi-source heterogeneous operation data includes electrical data, event data, state data and environmental data;

[0049] Step S2, data cleaning and time alignment are performed on the multi-source heterogeneous operation data, and normalized to obtain regular time series data;

[0050] Step S3, based on the regularized time series data, a health degree comprehensive calculation model of the transformer area is constructed, and a health degree index of each transformer area is calculated;

[0051] Step S4, according to the numerical range of the health degree index of the transformer area, a hierarchical early warning information is generated;

[0052] Step S5, based on the historical voltage data and the topology structure information of the transformer area, a low voltage trend of each node of the transformer area in a future period is predicted;

[0053] Step S6, the early warning information, the multi-source heterogeneous operation data, the health degree index and the low voltage trend prediction result are stored and visually displayed.

[0054] In the step S1, the electrical data includes three-phase voltage, three-phase current and zero sequence current; the event data includes switch tripping signal and protection action event; the state data includes terminal online state; and the environmental data includes transformer oil temperature.

[0055] In the step S2, the data cleaning specifically includes: an adaptive weighted sliding window filtering algorithm is adopted, wherein each data point in the sliding window is given an adaptive weight, the adaptive weight is determined by a time decay factor and a mutation degree of the point data relative to the neighborhood data, and a final cleaned data value is a weighted average value of all data in the window.

[0056] The calculation formula of the adaptive weighted sliding window filtering algorithm is:

[0057]

[0058] wherein, represents a cleaned data value at time t, represents an original data value at time t+i, represents an adaptive weight coefficient, represents a relative position index in the sliding window, represents a sliding window size;

[0059] wherein, the calculation formula of the adaptive weight coefficient is:

[0060]

[0061] wherein, represents a time decay constant, represents a difference between adjacent time point data, represents an absolute value;

[0062] The window size 2N+1 is matched, and usually The setting of the window edge weight is decayed to a very small value, which can be set .

[0063] The time alignment in the step S2 specifically includes: using a dynamic time warping algorithm to perform time alignment on the multi-source data.

[0064] The dynamic time warping algorithm is a technique for measuring the similarity between two time series, particularly suitable for those time series that may have different speeds at different time points. DTW finds the best matching path by nonlinearly warping the time series to minimize the distance between them.

[0065] The specific steps include: defining a distance matrix: given two time series X and Y, containing n and m data points respectively, first construct an n x m distance matrix D, where element D[i][j] represents the Euclidean distance or other appropriate distance measure between the i-th point of X and the j-th point of Y.

[0066] Cumulative distance matrix: then define a cumulative distance matrix γ, where γ[i][j] represents the total distance of the shortest path from the starting point (0,0) to the point (i,j). This shortest path needs to satisfy the following constraints:

[0067] Monotonicity: the path can only move to the right or upwards.

[0068] Boundary conditions: the path must start from (0,0) and end at (n,m).

[0069] Continuity: each step on the path must be continuous, i.e. cannot skip any points.

[0070] Recursive calculation: fill the cumulative distance matrix γ using dynamic programming method.

[0071] Get the optimal path: once the cumulative distance matrix is filled, you can start from γ[n][m] and gradually find the optimal path according to the principle of minimum cumulative distance by backtracking.

[0072] Normalize the distance: in order to compare time series of different lengths, the final minimum cumulative distance is usually normalized by dividing by some function of the path length.

[0073] In the step S3, the calculation method of the transformer area health degree comprehensive calculation model is: constructing a health degree basic index and a penalty factor respectively, multiplying the health degree basic index and the penalty factor of each transformer area to obtain the final health degree index of each transformer area.

[0074] The health degree basic index includes: voltage quality index term, communication state index term, device event index term and load balancing index term.

[0075] The penalty factor includes: a temperature-related penalty item and a voltage-related penalty item, wherein the temperature-related penalty item is composed of a temperature penalty coefficient and a transformer oil temperature overlimit indication function, and the voltage-related penalty item is composed of a voltage penalty coefficient and a voltage quality indication function.

[0076] Among them, the calculation formula of the comprehensive calculation model of the substation health degree is:

[0077]

[0078] in, Indicates the health index of the Taiwan area. Represents the basic health index, represents the penalty factor;

[0079] The calculation formula of the basic health index is:

[0080]

[0081] in, represents the voltage quality index, Indicates the communication status index, Indicates the device event index, Indicates the load balancing index, 、 、 and They represent the weights of the voltage quality index, communication status index, device event index, and load balancing index respectively. The weights are dynamically calculated based on the volatility of the historical data of each indicator using the entropy weight method. The greater the volatility of the indicator, the higher its weight, indicating that the indicator is more capable of distinguishing the differences in the health of the substation. and represents a nonlinear transformation function, which is used to map the original indicator value to the contribution and suppress extreme values;

[0082] in, is a standard Sigmoid function, the voltage quality index The higher the better, so Sigmoid is used directly to map it to contribution. For Taking the inversion and then inputting the standard Sigmoid function can ensure that the more events there are, the lower the contribution.

[0083] The calculation formula of voltage quality index is:

[0084]

[0085] in, denotes the total number of voltage monitoring points in the transformer area, denotes the jth voltage monitoring point, denotes the voltage value measured by the jth monitoring point at time t, denotes the rated voltage of the power grid;

[0086] The communication state index is the ratio of the number of terminal devices in online state at time t to the total number of terminal devices configured in the transformer area.

[0087] The calculation formula of the device event index is:

[0088]

[0089] wherein, denotes the decay coefficient, used to control the decay speed of the event on the index, denotes the recent time window, usually set to 24 hours, denotes the total number of tripping events of the protection device in the last 24 hours, denotes the proportional coefficient, used to convert the amplitude of the zero-sequence current into the equivalent event intensity, denotes the average value of the zero-sequence current in the last 24 hours;

[0090] The index is an exponentially decaying function, the more serious the tripping event and the ground fault feature, the larger the value in the parentheses, and the smaller the device event index. When there is no event, the device event index is 1.

[0091] The calculation formula of the load balancing index is:

[0092]

[0093] wherein, denotes the three-phase current at the outlet side of the transformer at time t, denotes the average value function; the fractional part in the formula calculates the three-phase unbalance degree, which is 1 when the three phases are completely balanced, and the higher the unbalance degree, the lower the load balancing index.

[0094] The calculation formula of the penalty factor is:

[0095]

[0096] wherein, denotes the temperature penalty coefficient, usually set to 0.3, denotes the indicator function, the function value is 1 when the condition in the parentheses is true, and the function value is 0 when the condition is false, denotes the transformer oil temperature, Safety threshold value representing the transformer temperature, usually in the range 85°C ~ 95°C (for oil-immersed distribution transformers), Voltage penalty coefficient, usually in the range 0.2 ~ 0.5, Critical threshold value representing the voltage quality index, usually set to 0.7;

[0097] In the step S4, the specific process of generating the hierarchical early warning information is: four health degree index intervals are pre-set, when the health degree index is greater than or equal to 0.8 and less than or equal to 1.0, it corresponds to a first level normal early warning, when the health degree index is greater than or equal to 0.6 and less than 0.8, it corresponds to a second level attention early warning, when the health degree index is greater than or equal to 0.4 and less than 0.6, it corresponds to a third level abnormal early warning, and when the health degree index is greater than or equal to 0 and less than 0.4, it corresponds to a fourth level serious early warning; the system calculates the health degree index in real time and judges the interval to which it belongs, and automatically generates the early warning signal of the corresponding level.

[0098] According to the numerical range, different levels of early warning information are generated:

[0099] First level (normal): 0.8 ≤ ≤ 1.0 (green)

[0100] Second level (attention): 0.6 ≤ < 0.8 (yellow)

[0101] Third level (abnormal): 0.4 ≤ < 0.6 (orange)

[0102] Fourth level (serious): 0 ≤ < 0.4 (red)

[0103] In the step S5, the low voltage trend of each node in the future period of the transformer area is predicted, which specifically includes: a undirected graph structure constructed based on the physical topology connection relationship of the transformer area, and historical voltage time series data corresponding to each node in the graph as model input; a spatio-temporal graph convolution network model is used for prediction, the spatio-temporal graph convolution network model includes 3 cascaded spatio-temporal convolution blocks; each spatio-temporal convolution block is composed of a graph convolution layer in spatial dimension and a one-dimensional convolution layer in time dimension in series; after the 3 spatio-temporal convolution blocks are stacked, the final voltage prediction value sequence corresponding to each node in a specific period in the future is obtained through a fully connected output layer.

[0104] Model input:

[0105] Node features: historical voltage data of each node (transformer, electric meter) in the transformer area topology network.

[0106] Graph structure: adjacency matrix A constructed based on the physical wiring diagram of the transformer area.

[0107] In the spatiotemporal convolution block, the graph convolution layer adopts a spectral graph convolution method based on Chebyshev polynomial approximation, and the order K of the Chebyshev graph polynomial is set to 3; the one-dimensional convolution layer adopts a dilated causal convolution structure, and the convolution kernel size of the dilated causal convolution layer is set to 3, and its dilation coefficient is set to 1, 2, and 4 in the three blocks respectively.

[0108] Setting K to 3 means that the receptive field of each node covers its 3rd-order neighbors (i.e., itself, direct neighbors, and neighbors of neighbors), which is usually sufficient to capture most of the spatial influences in the region.

[0109] The expansion coefficient is 1, 2, 4: the temporal receptive field of the top layer is 1 + (3-1) 1 + (3-1) 2 + (3-1) 4 = 1+2+4+8=15 time steps, which can effectively cover a sufficiently long historical period.

[0110] like Figure 2 As shown, a low-voltage substation operation status intelligent monitoring system includes:

[0111] Data acquisition module, used to collect multi-source heterogeneous operation data from various intelligent terminal devices in the low-voltage area;

[0112] Data preprocessing module, used to perform adaptive weighted sliding window filtering algorithm to clean and align data;

[0113] Health calculation module, used to calculate the dynamically weighted health index of the substation;

[0114] Intelligent early warning module, used to generate graded early warning information based on health index;

[0115] The spatiotemporal prediction module is used to predict low voltage trends based on the spatiotemporal graph convolutional network model;

[0116] The data storage and visualization module is used to store all data and analysis results and provide a graphical human-computer interaction interface.

[0117] The health calculation module includes:

[0118] The entropy weight calculation unit is used to dynamically calculate the weight coefficients of voltage quality, communication status, device events, and load balancing indicators based on the historical data volatility of each indicator;

[0119] Nonlinear transformation unit with built-in Sigmoid function for nonlinear mapping of voltage quality index and equipment event index;

[0120] A penalty factor calculation unit is configured to calculate a multiplication penalty factor according to the transformer oil temperature overrun and the voltage quality index overrun.

[0121] As shown in Figure 3 The specific structure of the space-time graph convolution network model is shown in the following table:

[0122] The input layer inputs historical voltage sequence data and a substation topology adjacency matrix A.

[0123] Three convolution blocks, each of which contains two sub-layers, and the arrows represent data flow:

[0124] The graph convolution layer processes the input data and outputs spatial features.

[0125] The one-dimensional convolution layer has a kernel size of 3, and the expansion coefficients are d=1, d=2 and d=4 respectively; the layer receives spatial features and outputs space-time features.

[0126] The output of the third convolution block is connected to a fully connected layer.

[0127] The final output layer outputs a future voltage prediction sequence.

[0128] As shown in Figure 4 The substation low-voltage monitoring display platform displays substation low-voltage records and trend graphs.

[0129] The examples described in the present application are only used to describe the preferred embodiments of the present application, and do not limit the concept and scope of the present application; various modifications and improvements of the technical solutions of the present application made by the engineering and technical personnel in the art without departing from the design idea of the present application shall fall within the protection scope of the present application.

Claims

1. A low-voltage transformer area operation state intelligent monitoring method, characterized in that, The method comprises the following steps: Step S1, collecting multi-source heterogeneous operation data through a sensing device deployed in a low-voltage transformer area, wherein the multi-source heterogeneous operation data comprises electrical data, event data, state data and environmental data; Step S2, performing data cleaning and time alignment on the multi-source heterogeneous operation data, and normalizing to obtain regular time series data; Step S3, constructing a transformer area health degree comprehensive calculation model based on the regular time series data, and calculating a health degree index of each transformer area; Step S4, generating graded early warning information according to the numerical range of the transformer area health degree index; Step S5, predicting the low-voltage trend of each node of the transformer area in the future period based on historical voltage data and transformer area topology structure information; Step S6, storing and visually displaying the early warning information, multi-source heterogeneous operation data, health degree index and low-voltage trend prediction result.

2. The method of claim 1, wherein, In the step S1, the electrical data includes three-phase voltage, three-phase current and zero-sequence current; the event data includes switch tripping signal and protection action event; the state data includes terminal online state; and the environmental data includes transformer oil temperature.

3. The method of claim 2, wherein, In the step S2, the data cleaning specifically comprises: adopting an adaptive weighted sliding window filtering algorithm, wherein each data point in the sliding window is given an adaptive weight, and the adaptive weight is determined by a time decay factor and the mutation degree of the data point relative to the neighborhood data; and the final cleaned data value is the weighted average value of all data in the window.

4. The method of claim 3, wherein, In the step S2, the time alignment specifically comprises: adopting a dynamic time warping algorithm to perform time alignment on the multi-source data.

5. The method of claim 4, wherein, In the step S3, the calculation method of the transformer area health degree comprehensive calculation model is: a health degree basic index and a penalty factor are constructed respectively, and the health degree basic index and the penalty factor of each transformer area are multiplied to obtain the final health degree index of each transformer area; Wherein, the health degree basic index comprises: a voltage quality index term, a communication state index term, a device event index term and a load balancing index term; The penalty factor comprises: a temperature related penalty term and a voltage related penalty term, wherein the temperature related penalty term is composed of a temperature penalty coefficient and a transformer oil temperature out-of-limit indicator function, and the voltage related penalty term is composed of a voltage penalty coefficient and a voltage quality indicator function.

6. The method of claim 5, wherein, In the step S4, the specific process of generating graded early warning information is: four health degree index intervals are preset, when the health degree index is greater than or equal to 0.8 and less than or equal to 1.0, it corresponds to a first level normal early warning, when the health degree index is greater than or equal to 0.6 and less than 0.8, it corresponds to a second level attention early warning, when the health degree index is greater than or equal to 0.4 and less than 0.6, it corresponds to a third level abnormal early warning, and when the health degree index is greater than or equal to 0 and less than 0.4, it corresponds to a fourth level serious early warning; the system calculates the health degree index in real time and judges its belonging interval, and automatically generates the early warning signal of the corresponding level.

7. The method of claim 6, wherein, The step S5 includes: constructing a undirected graph structure based on the physical topology connection relationship of the substation, and using the historical voltage time series data corresponding to each node in the graph as model input; predicting the low voltage trend of each node in the substation in the future period by using a spatio-temporal graph convolution network model, the spatio-temporal graph convolution network model includes three cascaded spatio-temporal convolution blocks; each spatio-temporal convolution block is composed of a graph convolution layer in spatial dimension and a one-dimensional convolution layer in time dimension in series; after the three spatio-temporal convolution blocks are stacked, the final voltage prediction value sequence corresponding to each node in a specific period in the future is obtained through a fully connected output layer.

8. The method of claim 7, wherein, In the spatio-temporal convolution block, the graph convolution layer adopts a spectral graph convolution method based on Chebyshev polynomials, and the order K of the Chebyshev graph polynomial is set to 3; the one-dimensional convolution layer adopts a dilated causal convolution structure, and the convolution kernel size of the dilated causal convolution layer is set to 3, and the dilated coefficients are set to 1, 2 and 4 in the three blocks respectively.

9. A low-voltage transformer area operation state intelligent monitoring system, characterized in that, Comprise: A data acquisition module for collecting multi-source heterogeneous operation data from various intelligent terminal devices in the low-voltage substation; A data preprocessing module for performing an adaptive weighted sliding window filtering algorithm to clean and align the data; A health degree calculation module for calculating a dynamically weighted substation health degree index; An intelligent early warning module for generating graded early warning information according to the health degree index; A spatio-temporal prediction module for predicting the low voltage trend based on a spatio-temporal graph convolution network model; A data storage and visualization module for storing all data and analysis results, and providing a graphical human-computer interaction interface.

10. The system of claim 9, wherein, The health degree calculation module comprises: An entropy weight method calculation unit for dynamically calculating the weight coefficients of the voltage quality, communication status, device event and load balancing indexes according to the historical data volatility of each index; A nonlinear transformation unit with a Sigmoid function for nonlinear mapping processing of the voltage quality index and the device event index; A penalty factor calculation unit for calculating a multiplicative penalty factor according to the transformer oil temperature overrun and the voltage quality index overrun.

Images