Transformer substation hot spot prediction method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses a transformer substation hot spot prediction method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining time data and spatial data of a transformer substation component array in a current time period; according to the first sequence, the second sequence and a preset physical feature constraint, a dynamic space-time diagram of the transformer substation component array in the current time period is constructed, and the dynamic space-time diagram is used for reflecting a first dynamic change condition of hot spots on the transformer substation component array in the current time period; according to the dynamic time-space diagram, predicting a hot spot probability diagram and a hot spot evolution direction vector of the transformer substation component array at a target moment, the target moment being later than the current moment; and according to the hot spot probability graph and the hot spot evolution direction vector, predicting a second dynamic change condition of the hot spots of the substation component array from the current moment to the target moment. By adopting the embodiment of the invention, the fault risk of the transformer substation can be reduced in the operation process of the transformer substation.

Description

Technical Field

[0001] The present invention relates to the technical field of substations, and in particular to a method, device, electronic equipment and storage medium for predicting hot spots in substations. Background Art

[0002] In recent years, hot spot defects have become a key factor affecting the performance and lifespan of substation components during operation and maintenance. Common methods for detecting hot spots include direct measurement of component surface temperature, capturing hot spot images using infrared thermal imaging, and indirectly identifying the presence of hot spots based on abnormal changes in electrical parameters.

[0003] However, temperature measurement and infrared thermal imaging technologies mainly focus on detecting whether hot spots have appeared in substation components in the past. At best, they can only detect whether hot spots exist at the current moment in real time. They cannot determine the evolution trend of hot spots at different time points in the future. As a result, operation and maintenance personnel are unable to formulate effective prevention and response measures in advance, increasing the risk of substation failure. Summary of the Invention

[0004] In order to solve the above problems, embodiments of the present invention provide a method, device, electronic device and storage medium for predicting hot spots in a substation, which can reduce the failure risk of the substation during operation.

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting hot spots in a substation, comprising:

[0006] Acquiring time data and spatial data of a substation component array in a current time period, wherein the time data includes a first sequence corresponding to a first parameter, the first sequence being a sequence consisting of a change in the first parameter over time in the current time period, the first parameter including at least one of the following: temperature, electrical parameters, and environmental data of the substation component array; and the spatial data includes a second sequence corresponding to a second parameter, the second sequence being a sequence consisting of a change in the second parameter over time in the current time period, the second parameter including at least one of the following: spatial locations of hot spots and shadow areas of the substation component array;

[0007] Constructing a dynamic spatiotemporal graph of the substation component array in a current time period based on the first sequence, the second sequence, and preset physical feature constraints, wherein the dynamic spatiotemporal graph is used to reflect a first dynamic change of hot spots on the substation component array in the current time period;

[0008] Based on the dynamic spatiotemporal graph, the hot spot probability map and hot spot evolution direction vector of the substation component array at the target time are predicted, where the target time is later than the current time;

[0009] According to the hot spot probability map and the hot spot evolution direction vector, the second dynamic change of the hot spots of the substation component array from the current moment to the target moment is predicted.

[0010] In a second aspect, an embodiment of the present invention provides a substation hot spot prediction device, comprising an acquisition unit and a processing unit;

[0011] an acquisition unit, configured to acquire time data and spatial data of a substation component array in a current time period, wherein the time data includes a first sequence corresponding to a first parameter, the first sequence being a sequence consisting of a change in the first parameter over time in the current time period, the first parameter including at least one of the following: temperature, electrical parameters, and environmental data of the substation component array; and the spatial data includes a second sequence corresponding to a second parameter, the second sequence being a sequence consisting of a change in the second parameter over time in the current time period, the second parameter including at least one of the following: spatial locations of hot spots and shadow areas of the substation component array;

[0012] a processing unit, configured to construct a dynamic spatiotemporal graph of the substation component array in a current time period based on the first sequence, the second sequence, and preset physical characteristic constraints, wherein the dynamic spatiotemporal graph is configured to reflect a first dynamic change of hot spots on the substation component array in the current time period;

[0013] Based on the dynamic spatiotemporal graph, the hot spot probability map and hot spot evolution direction vector of the substation component array at the target time are predicted, where the target time is later than the current time;

[0014] According to the hot spot probability map and the hot spot evolution direction vector, the second dynamic change of the hot spots of the substation component array from the current moment to the target moment is predicted.

[0015] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a processor and a memory, wherein the processor is connected to the memory, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory so that the electronic device performs the method described in the first aspect.

[0016] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method as described in the first aspect.

[0017] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer is operable to enable the computer to execute the method described in the first aspect.

[0018] The implementation of the embodiments of the present application has the following beneficial effects:

[0019] In the embodiment of the present application, the time data and spatial data of the substation component array in the current time period are first obtained, wherein the time data includes a first sequence corresponding to a first parameter, the first sequence is a sequence consisting of the variation of the first parameter over time in the current time period, and the first parameter includes at least one of the following: the temperature, electrical parameters and environmental data of the substation component array; the spatial data includes a second sequence corresponding to a second parameter, the second sequence is a sequence consisting of the variation of the second parameter over time in the current time period, and the second parameter includes at least one of the following: the spatial position of the hot spot and the shadow area of ​​the substation component array; then, according to the first sequence, The dynamic space-time diagram of the substation component array in the current time period is constructed based on the first sequence, the second sequence and the preset physical feature constraints, wherein the dynamic space-time diagram is used to reflect the first dynamic change of the hot spots on the substation component array in the current time period. Next, based on the dynamic space-time diagram, the hot spot probability map and the hot spot evolution direction vector of the substation component array at the target moment are predicted, wherein if the target moment is later than the current moment, the target moment can be understood as the future moment relative to the current moment. Finally, based on the hot spot probability map and the hot spot evolution direction vector, the second dynamic change of the hot spots of the substation component array from the current moment to the target moment is predicted. Therefore, by obtaining the time data and spatial data of the substation component array in the current time period, and predicting the hot spot probability map and hot spot evolution direction vector of the substation component array at the future moment based on the dynamic time-space diagram, the second dynamic change of the hot spot of the substation component array from the current moment to the target moment can be predicted based on the hot spot probability map and the hot spot evolution direction vector, thereby being able to predict the dynamic change of the hot spot in time and space from the current moment to the future moment, and then the operation and maintenance personnel can formulate effective prevention and response measures in advance based on the dynamic change, thereby reducing the failure risk of the substation. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the background technology, the drawings required for use in the embodiments of the present invention or the background technology will be described below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 This is a schematic diagram of the architecture of a substation hot spot prediction system provided by an embodiment of the present application;

[0022] Figure 2 This is a flow chart of a method for predicting hot spots in substations provided by an embodiment of the present application;

[0023] Figure 3This is a schematic diagram of a dynamic space-time graph provided in an embodiment of the present application;

[0024] Figure 4 This is a schematic diagram of the structure of a hot spot defect prediction model provided in an embodiment of the present application;

[0025] Figure 5 Schematic diagram of a hot spot probability map provided in an embodiment of the present application;

[0026] Figure 6 Schematic diagram of a hot spot evolution direction vector provided in an embodiment of the present application;

[0027] Figure 7 This is a schematic structural diagram of a substation hot spot prediction device provided by an embodiment of the present application;

[0028] Figure 8 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0030] The terms "first," "second," "third," and "fourth," etc., in the specification and claims of this application and the accompanying drawings are used to distinguish between different objects, rather than to describe a particular order. In addition, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or modules is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other steps or modules inherent to the process, method, product, or apparatus.

[0031] References herein to "embodiments" mean that a particular feature, result, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0032] The following describes the relevant contents, concepts, technical issues, technical solutions, beneficial effects, etc. involved in the embodiments of this application.

[0033] See Figure 1 , Figure 1 The following is a schematic diagram of the architecture of a substation hot spot prediction system provided in an embodiment of the present application. The substation hot spot prediction system includes a substation and a server. The substation is provided with a substation component array. The component array in the substation is a structured unit formed by arranging and combining various types of electrical equipment according to a certain pattern to achieve functions such as power conversion, distribution, and control. Data is exchanged between the substation and the server. For example, the server sends a data acquisition request to the substation, and the substation sends the time data and spatial data of the substation component array in the current time period to the server based on the data acquisition request received from the server.

[0034] See Figure 2 , Figure 2 This is a flow chart of a method for predicting hot spots in substations provided by an embodiment of the present application. The method for predicting hot spots in substations provided by an embodiment of the present application includes but is not limited to the following steps:

[0035] Step S101: obtaining time data and spatial data of a substation component array in a current time period;

[0036] The time data includes a first sequence corresponding to a first parameter, the first sequence being a sequence consisting of changes in the first parameter over time within a current time period, the first parameter including at least one of the following: temperature, electrical parameters, and environmental data of a substation component array; the spatial data includes a second sequence corresponding to a second parameter, the second sequence being a sequence consisting of changes in the second parameter over time within the current time period, the second parameter including at least one of the following: spatial locations of hot spots and shadow areas of the substation component array;

[0037] Step S102: constructing a dynamic spatiotemporal graph of the substation component array in the current time period according to the first sequence, the second sequence and preset physical feature constraints;

[0038] The dynamic spatiotemporal graph is used to reflect the first dynamic change of the hot spots on the substation component array within the current time period;

[0039] Step S103: predicting the hot spot probability map and hot spot evolution direction vector of the substation component array at the target time based on the dynamic space-time map;

[0040] Among them, the target time is later than the current time;

[0041] Step S104: predicting a second dynamic change of the hot spots of the substation component array from the current moment to the target moment based on the hot spot probability map and the hot spot evolution direction vector.

[0042] Specifically, substation operation is a complex spatiotemporal process. The generation, development, and evolution of hot spot defects are closely related to temporal factors, such as light intensity and ambient temperature fluctuations at different times, which affect the temperature distribution and electrical performance of substation components. They are also influenced by spatial factors, such as the position of substation components in the array and shadow distribution, which can lead to local temperature differences and current unevenness, thus causing hot spots. Therefore, comprehensive temporal and spatial data is required to accurately describe the substation's operating status and the potential conditions for the occurrence of hot spot defects.

[0043] Specifically, the current time period is the time period between the historical moment and the current moment, the historical moment is earlier than the current moment, the time data includes real-time temperature field, electrical parameter flow and environmental time series data, the spatial data includes the daily detection data of hot spots on the back of the substation component array and the regional shadow dynamic modeling data, the regional shadow dynamic modeling data is used to reflect the position and area of ​​the hot spot, wherein the real-time temperature field includes the temperature data and deformation data of the substation component array in the current time period, the deformation data is used to reflect the component deformation data, the substation component array will be deformed due to external loads, material aging and installation errors, the electrical parameter flow includes the current-voltage relationship curve of the substation component array junction box in the current time period, the environmental time series data includes the irradiance accumulation data, wind speed accumulation data and dust accumulation data of the substation component array in the current time period.

[0044] Specifically, substation operation adheres to certain physical laws, such as heat conduction, power loss, and shadow effects. Thermal resistance constraints describe the degree of resistance to heat transfer and temperature distribution within substation components. Power loss constraints reflect the energy loss of substation components during power generation. Shadow composite effect constraints account for the impact of shadows on substation component performance and hot spot distribution. Incorporating these physical constraints into the construction of dynamic spatiotemporal graphs can make them more consistent with actual substation operation.

[0045] Specifically, the preset physical characteristic constraints include thermal resistance constraint, power loss constraint and shadow composite effect constraint. Among them, the thermal resistance constraint is used to describe the degree of obstruction of heat transfer inside the substation components and the temperature distribution of the substation components. The power loss constraint is used to describe the power loss of the substation components during the power generation process. The shadow composite effect constraint is used to assist in predicting the distribution and changes of hot spots in the shadow area.

[0046] Furthermore, the hot spot probability map intuitively displays the probability of hot spots appearing at different times and spatial locations. Operations and maintenance personnel can quickly understand the areas and times where hot spots are likely to occur by viewing the probability map. The vector representing the hot spot's evolution direction provides information on the hot spot's development trend, helping to predict the hot spot's migration path and diffusion range. By outputting these two results, comprehensive and detailed hot spot prediction information can be provided to operations and maintenance personnel, enabling them to formulate targeted maintenance strategies in advance, such as focusing on monitoring areas where hot spots are likely to appear and promptly cleaning dust that affects heat dissipation. This can reduce the damage caused by hot spots to substation components and improve the substation's operating efficiency and stability.

[0047] In one possible embodiment, based on the hot spot probability map and the hot spot evolution direction vector representing the hot spot evolution direction, the causes of the hot spots are decomposed and a three-dimensional hot spot evolution sandbox is constructed to obtain the second dynamic change of the hot spot defect in time and space. Exemplarily, for the cause of hot spots, dust, bird droppings, fallen leaves, cloud shadows and other causes may block part of the battery cell, resulting in current mismatch. The blocked area appears as a high-probability hot spot in the probability map and may expand over time, such as dust accumulation. The hot spots caused by the blockage usually extend along the edge of the blockage, and the vector direction is consistent with the movement path of the blockage.

[0048] For example, based on the occurrence probability of hot spots at different time and space positions, a probability-color temperature nonlinear mapping relationship is constructed to convert the hot spot probability map into a dynamic heat map with a color temperature gradient; the vector representing the evolution direction of the hot spot is normalized to generate a vector heat map with directional field characteristics, wherein the arrow length is used to characterize the hot spot diffusion intensity after logarithmic transformation, and the arrow direction is used to indicate the hot spot migration path. The hue component in the HSV (Hue, Saturation, Value) color space is used to encode the time evolution dimension, wherein the HSV color space is a color space created according to the intuitive characteristics of color, also known as the hexagonal pyramid model, each color is represented by hue (Hue, H), saturation (S) and value (Value, V), and the color parameters in this model are: hue (H), saturation (S), brightness (V); according to the Global Positioning System (GPS) of the substation component array A three-dimensional substation component array sand table model with terrain adaptation characteristics is constructed using a digital elevation model based on the geographic coordinates, azimuth, and dual-axis tilt parameters of the GPS System. Dynamic thermal maps and vector thermal maps are integrated into a two-dimensional geographic layout map of the substation component array, which is then mapped into a physically based rendering (PBR) material base texture on the top surface of the three-dimensional substation component array sand table model. The infrared radiation field on the surface of the three-dimensional substation component array sand table model is rendered in real time using a ray tracing algorithm to simulate the multispectral characteristics of a real thermal imager. Based on the multispectral characteristics, a spatiotemporally continuous hot spot evolution sequence is generated along the time dimension, and an interpolation algorithm is used to construct a three-dimensional hot spot evolution sand table with variable speed playback to obtain the dynamic changes of hot spot defects in time and space.

[0049] In the embodiments of this application, a three-dimensional hot spot evolution sandbox displays the dynamic changes of hot spot defects in time and space in an intuitive and dynamic manner. By observing the sandbox, operations and maintenance personnel can gain a deeper understanding of the causes, development trends, and evolution patterns of hot spots, such as how hot spots propagate between different components and how the diffusion intensity changes. This visual display method helps operations and maintenance personnel better understand the hot spot problem, make more scientific and reasonable operations and maintenance decisions, improve the operation and maintenance efficiency and management level of substations, and reduce power generation losses and equipment damage caused by hot spots.

[0050] Optionally, the substation component array includes n components; step S102, constructing a dynamic spatiotemporal graph of the substation component array in the current time period according to the first sequence, the second sequence, and preset physical feature constraints, may include the following steps:

[0051] Step S201: determining the node characteristics of each component according to the first sequence, wherein the node characteristics of each component are used to reflect the temperature gradient, electrical parameters and aging index of each component in the current time period;

[0052] Step S202: determining the shadow propagation paths between n components according to the second sequence;

[0053] Step S203: Determine the adjacency relationship between n components according to the shadow propagation path;

[0054] Step S204: determining edge features between adjacent components among the n components based on the adjacency relationship, wherein the edge features between adjacent components are used to reflect the distance, temperature gradient, and electrical parameter correlation between adjacent components in the current time period;

[0055] Step S205: determining edge weights between adjacent components based on edge features and physical feature constraints between adjacent components;

[0056] Step S206: Construct a dynamic spatiotemporal graph based on the n node features of the n components and the edge weights between adjacent components.

[0057] Specifically, the substation component array includes n components, where n is a positive integer. The n components can be a transformer array or a converter array for energy conversion, a circuit breaker array or a disconnector array for switching and control, or a mutual inductor array or a relay protection device array for protection and measurement, without limitation.

[0058] In one possible embodiment, node features for each component are determined based on the first sequence, where the node features of each component are used to reflect the temperature gradient, electrical parameters, and aging index of each component within the current time period. Specifically, the temperature gradient of the substation component is calculated based on the temperature and deformation data of the substation component array; the electrical parameters of the substation component are obtained based on the current-voltage relationship curve of the substation component array junction box; and the aging index of the substation component is calculated based on the accumulated irradiance data, wind speed data, and dust accumulation data of the substation component array. Each substation component is treated as a node in a dynamic spatiotemporal graph, and the temperature gradient, electrical parameters, and aging index of the substation component are used as node features.

[0059] In one possible embodiment, the shadow propagation path between n components is determined based on the second sequence. First, based on the data related to the shadow propagation path in the second sequence, for example, satellite imagery for identifying static shadows, real-time infrared thermal images for detecting dynamic shadows, and an irradiance sensor network for providing shadow intensity data, a spatiotemporal continuity constraint is constructed. In the temporal dimension, the shadow regions of adjacent frames must satisfy an intersection-over-union (IOU) threshold greater than a preset IOU threshold. In the spatial dimension, the shadow boundary smoothness constraint must be less than or equal to a preset smoothness threshold. Next, based on a fluid dynamics propagation model, the shadow is considered as a cold fluid that propagates in the opposite direction of the temperature gradient. The propagation speed is the product of the temperature gradient and a preset propagation coefficient. The preset propagation coefficient can be 0.01, 0.02, 0.05, etc. A state transition matrix is ​​constructed, an initial state vector is set, and the probability of the shadow propagating from component i to component j at the first moment is determined through iterative calculation. Based on this probability, the shadow distribution at the first moment is obtained.

[0060] In one possible embodiment, the adjacency relationship between n components is determined based on the shadow propagation path. An electrical connection matrix is ​​constructed based on the wiring diagram. If components i and j are directly electrically connected, such as being connected in series on the same branch, then there is an adjacency relationship between components i and j. Finite element simulation calculations also determine if the impact of the thermal change of component i on j exceeds a threshold, such as a temperature rise of ≥0.5°C, then there is an adjacency relationship between components i and j.

[0061] In one possible embodiment, edge features between adjacent components among n components are determined based on the adjacency relationship, where the edge features between adjacent components are used to reflect the distance, temperature gradient, and electrical parameter correlation between adjacent components within the current time period. The distance between adjacent components can be obtained by directly calculating the distance between the geometric centers of the components, or by considering the thermal conductivity of the materials and calculating the equivalent thermal resistance to obtain the thermal resistance distance. The thermal resistance distance is used as the distance between adjacent components. The equivalent thermal resistance between adjacent components = component geometric center distance / (thermal conductivity × contact area between adjacent components).

[0062] In a possible embodiment, the edge weights between adjacent components are determined based on the edge features and physical feature constraints between adjacent components. The first constraint parameter is calculated based on the thermal conduction constraint in the physical feature constraint, the second constraint parameter is calculated based on the electrical constraint in the physical feature constraint, the third constraint parameter is determined based on whether the adjacent component is in a hot spot, and the first weight, the second weight and the third weight are determined, edge weight = first weight × first constraint parameter + second weight × second constraint parameter + third weight × third constraint parameter. Exemplarily, the first constraint parameter is determined based on the equivalent thermal resistance, reference thermal resistance and direction influence factor, the second constraint parameter is determined based on the voltage value and correlation coefficient influence factor of the adjacent component, and different third constraint parameters are determined based on whether one of the adjacent components is in a hot spot, or both of the adjacent components are in a hot spot, or whether none of the adjacent components are in a hot spot.

[0063] In one possible embodiment, a dynamic spatiotemporal graph is constructed based on n node features of n components and edge weights between adjacent components. For example, each substation component is used as a node in the dynamic spatiotemporal graph, with the temperature gradient, electrical parameters, and aging index of the substation component as node features, the shadow propagation path as an edge of the dynamic spatiotemporal graph, and the spacing, temperature gradient, and electrical parameter correlation between substation components as edge features.

[0064] Furthermore, the temporal and spatial data of the substation component array are used as input features of the Long Short-Term Memory (LSTM) network to update the graph structure of the spatiotemporal graph in real time and perform adaptive adjustments.

[0065] In the specific embodiment, see Figure 3 , Figure 3 This is a schematic diagram of a dynamic spatiotemporal graph provided by an embodiment of the present application. If the substation component array includes five components, A, B, C, D, and E, in the dynamic spatiotemporal graph, each component is a node, each node includes the node characteristics of the component corresponding to the node, and adjacent components are characterized by edge weights. For example, the node characteristic of component A is node characteristic A, and the edge weight between components B and C is edge weight BC. The node characteristics of each component and the edge weights between adjacent components are shown in the figure and will not be repeated here.

[0066] Optionally, step S103, predicting the hot spot probability map and hot spot evolution direction vector of the substation component array at the target time based on the dynamic spatiotemporal graph, may include the following steps:

[0067] Step S301: Determine the time series features and spatial series features corresponding to each component based on the dynamic spatiotemporal graph, wherein the time series features are used to reflect the temporal propagation pattern of the hot spot in each component, and the spatial series features are used to reflect the spatial propagation pattern of the hot spot in each component;

[0068] Step S302: Determine a remote heat propagation effect characteristic representation of each component based on the time series characteristics and spatial series characteristics corresponding to each component, wherein the remote heat propagation effect characteristic representation of each component is used to reflect the diffusion speed, diffusion direction, and diffusion distance of the hot spot corresponding to each component;

[0069] Step S303: predicting the hot spot probability of each component at the target time based on the long-range heat propagation effect characteristic representation of each component;

[0070] Step S304: Determine a hot spot probability map and a hot spot evolution direction vector based on the n hot spot probabilities of the n components.

[0071] In a specific embodiment, a hot spot defect prediction model is used to predict the hot spot probability map and hot spot evolution direction vector of the substation component array at the target time. Figure 4 , Figure 4 This is a structural diagram of a hot spot defect prediction model provided in an embodiment of the present application. The hot spot defect prediction model includes: an input layer, an LSTM network layer enhanced by physical information, a graph neural network layer guided by an attention mechanism, a probabilistic space-time capsule network layer, and an output layer. Specifically, the input layer is used to receive the dynamic spatiotemporal graph and extract time series features and spatial series features, and input them into the LSTM network layer enhanced with physical information; the LSTM network layer enhanced with physical information is used to receive time series features and spatial series features, and use the heat conduction equation as a regularization constraint to guide the LSTM network to learn features that both conform to physical laws and have a correlation with hot spots exceeding a preset threshold, and input them into the graph neural network layer guided by the attention mechanism; the graph neural network layer guided by the attention mechanism is used to use the attention mechanism to capture the feature representation of the remote heat propagation effect of time series features and spatial series features that conform to physical laws and have a correlation with hot spots exceeding a preset threshold; the probabilistic space-time capsule network layer is used to predict the probability of occurrence of hot spots at different time and space positions based on the feature representation of the remote heat propagation effect, and model the probability of occurrence of hot spots at different time and space positions through the capsule structure to obtain a hot spot probability map containing the probability of occurrence of hot spots at different time and space positions and a vector representing the evolution direction of the hot spots; the output layer is used to output a hot spot probability map containing the probability of occurrence of hot spots at different time and space positions and a vector representing the evolution direction of the hot spots.

[0072] For example, see Figure 5 , Figure 5This is a schematic diagram of a hot spot probability map provided by an embodiment of the present application. If the substation component array includes five components A, B, C, D, and E, and at the target time, if the hot spot probabilities of the five components are hot spot probability A, hot spot probability B, hot spot probability C, hot spot probability D, and hot spot probability E, then the hot spot probability map generated based on the five hot spot probabilities can be as follows: Figure 5 shown.

[0073] For example, see Figure 6 , Figure 6 This is a schematic diagram of a hot spot evolution direction vector provided by an embodiment of the present application. If the substation component array includes five components A, B, C, D, and E, the hot spot evolution direction vector can be obtained by Figure 6 The data table shown is used for representation, and the size of the numerical value in the table represents the size of the hot spot evolution direction vector between adjacent components, which is used to indicate the hot spot evolution intensity. The positive and negative of the numerical value in the table represents the direction of the hot spot evolution direction vector between adjacent components. For example, at the current moment, there are hot spots in component A and component B, and no hot spots in other components. According to the table, it can be obtained that there are hot spots in component A and component C at the target moment. Since the corresponding value between AA is 1, it indicates that from the current moment to the target moment, the hot spot at component A will still evolve to component A. Since the corresponding value between BA is 2, and the corresponding value between AB is -2, and the corresponding value between BB is 0, it indicates that from the current moment to the target moment, the hot spot at component B will evolve to component A, and the evolution intensity is higher. Since the corresponding value between BC is 1, and the corresponding value between CB is -1, it indicates that from the current moment to the target moment, the hot spot at component B will also evolve to component C, and the evolution intensity is lower than the intensity of the evolution from component B to component A.

[0074] Optionally, step S301, determining the time series features and space series features corresponding to each component according to the dynamic spatiotemporal graph, may include the following steps:

[0075] Step S401: performing time series analysis on the node features of each component according to the dynamic spatiotemporal graph to obtain m subsequences corresponding to each component;

[0076] Step S402: extract features from the m subsequences corresponding to each component to obtain the time series features corresponding to each component;

[0077] Step S403: constructing a spatial adjacency matrix based on edge features and edge weights between adjacent components;

[0078] Step S404: According to the preset graph convolutional network, the spatial adjacency matrix and the edge features between adjacent components are processed to obtain the spatial sequence features of each component.

[0079] Specifically, based on the dynamic spatiotemporal graph, a time series analysis is performed on the node features of each component, resulting in m subsequences corresponding to each component, where m is a positive integer. First, the node features of each component are sorted by timestamp to form the original time series. Then, the window length and step size are set. The window length can be 24 hours, corresponding to 96 15-minute sampling points, and the step size can be 1 hour, i.e., 4 sampling points. This ensures overlap between subsequences to preserve temporal continuity. Both short and long windows are applied. The short window can be 1 hour to capture rapidly changing features, and the long window can be 72 hours to capture periodic changes, forming a set of subsequences with different time granularities. In addition, each subsequence is Z-score normalized, replacing outliers that deviate from the mean by more than 3 standard deviations with the weighted average of the adjacent valid values.

[0080] In one possible embodiment, feature extraction is performed on the m subsequences corresponding to each component to obtain the time series features corresponding to each component. The feature extraction includes statistical feature extraction, time domain feature extraction, and frequency domain feature extraction. Specifically, the mean, median, standard deviation, skewness, kurtosis, etc. of each subsequence are calculated to quantify the data distribution characteristics and perform statistical feature extraction. The rate of change and acceleration of adjacent points are calculated by a differential operator to identify sudden changes such as temperature rise or current fluctuations and perform time domain feature extraction. The subsequences are converted to the frequency domain using a fast Fourier transform to extract the main frequency component, harmonic energy distribution, and periodic intensity, and perform frequency domain feature extraction.

[0081] In one possible embodiment, a spatial adjacency matrix is ​​constructed based on the edge features and edge weights between adjacent components. Multidimensional edge features, such as the physical distance between components, thermal conductivity, and electrical connection strength, are normalized to integrate the edge features. Edge weights are then calculated based on a physical constraint model. The temperature gradient influencing factor is derived using the heat conduction equation, and the electrical parameter correlation is calculated using a dynamic time warping algorithm to measure curve similarity. An n×n matrix is ​​initialized, with the matrix elements representing the edge weights between corresponding components. Weights between non-adjacent components are set to minimum values ​​to preserve weak connections.

[0082] In one possible embodiment, a pre-defined graph convolutional network is used to process the spatial adjacency matrix and edge features between adjacent components to obtain spatial sequence features for each component. A three-layer structure is employed: the first layer performs feature transformation, the middle layer captures multi-hop neighbor information, and the final layer outputs a spatial embedding representation. Adjacency matrix-weighted aggregation of adjacent node features is used, with edge features serving as message passing weights to enhance the expressiveness of spatial relationships. Furthermore, a self-attention mechanism is used to calculate the attention weights of nodes towards different neighbors. A residual structure is used to address the vanishing gradient problem in deep networks, and batch normalization is used to accelerate model convergence and improve stability.

[0083] In a specific embodiment, node features of a dynamic spatiotemporal graph are received and time series analysis is performed, and the node features of the dynamic spatiotemporal graph are divided into multiple subsequences using a sliding window technique; feature extraction is performed on each subsequence using a statistical method, and frequency domain features are extracted using Fourier transform as time series features; spatial sequence analysis is performed on edge features of the dynamic spatiotemporal graph, and a spatial adjacency matrix is ​​constructed based on edge features and edge weights; the spatial adjacency matrix and edge features are processed using a graph convolutional network to extract spatial sequence features that reflect spatial position relationships.

[0084] In the embodiment of the present application, dynamic slicing of time series data is achieved through sliding windows to adapt to the time-varying characteristics of component states, and the multi-scale window design takes into account both short-term mutations and long-term trends.

[0085] Optionally, step S302, determining the remote heat propagation effect characteristic representation of each component based on the time series characteristics and spatial series characteristics corresponding to each component, may include the following steps:

[0086] Step S501: performing a first dot product calculation on the time series features between adjacent components based on the time series features and adjacency relationship corresponding to each component to obtain a first dot product calculation result between the adjacent components;

[0087] Step S502: performing a second dot product calculation on the spatial sequence features between adjacent components based on the spatial sequence features and adjacency relationship corresponding to each component to obtain a second dot product calculation result between the adjacent components;

[0088] Step S503: determining a remote heat propagation effect characteristic representation of each component based on the first dot product calculation result between adjacent components and the second dot product calculation result between adjacent components.

[0089] In one possible embodiment, based on the time series features and adjacency relationships corresponding to each component, a first dot product calculation is performed on the time series features between adjacent components to obtain the first dot product calculation results between adjacent components. The time series features of adjacent components are aligned using a dynamic time warping algorithm to eliminate feature misalignment caused by time offset. For the aligned time series features, dot products are calculated by dimension, such as the dot product of the temperature dimension, the dot product of the current dimension, etc., to obtain a multi-dimensional dot product result vector. Moreover, for each component, not only is the dot product with the directly adjacent components calculated, but the dot product between remote components is also calculated through hop count constraints to capture indirect heat propagation relationships. The calculation results are weighted using the heat conduction coefficient. The closer the distance and the stronger the heat conduction capability of the component pair, the higher the weight of the dot product result.

[0090] In one possible embodiment, a second dot product calculation is performed on the spatial sequence features between adjacent components based on the spatial sequence features and adjacency relationships corresponding to each component, thereby obtaining the second dot product calculation results between adjacent components. The high-dimensional spatial sequence features are projected into a unified feature space through a linear transformation to ensure that different types of features, such as distance, thermal resistance, and electrical connection strength, can be processed. A directional factor is introduced into the dot product calculation to give a higher weight to the influence of upstream components on downstream components, simulating the directionality of heat conduction. Both local and global spatial relationships are considered, where local spatial relationships refer to directly adjacent components, and global spatial relationships refer to network-level associations captured through the graph Laplacian matrix. Exponential decay is applied to historical dot product results to make recent spatial relationships have a greater impact on current attention.

[0091] In one possible embodiment, a characteristic representation of the remote heat propagation effect of each component is determined based on the results of the first dot product calculation between adjacent components and the second dot product calculation between adjacent components. The time series dot product results and the spatial series dot product results are weighted and fused. The weight coefficients are learned through gradient descent to reflect the relative importance of temporal and spatial information to heat propagation. The fused attention scores are normalized using a softmax function, but a physical constraint term is added before the exponential operation to ensure that paths that violate the heat conduction law receive extremely low weights. Based on the normalized attention weights, the time series features and spatial series features of adjacent components are weightedly summed, while taking into account the contributions of remote components. A nonlinear transformation is applied to the aggregated results to highlight the characteristics of key heat propagation paths and suppress noisy paths.

[0092] In a specific embodiment, a dot product attention mechanism is used to calculate the dot product between the time series features of each pair of nodes and the dot product between the spatial sequence features of the edges between each pair of nodes to obtain an attention score; the attention score is normalized using a softmax function to obtain a normalized attention weight; according to the normalized attention weight, the time series features of the nodes and the spatial sequence features of the edges are weighted and summed respectively to obtain a characteristic representation of the long-range heat propagation effect of the time series features and spatial sequence features that conform to physical laws and whose correlation with the hot spot exceeds a preset threshold.

[0093] In the embodiment of the present application, by calculating the dot product of the time series and the space series separately and then performing cross-modal fusion, collaborative modeling of the spatiotemporal dimensions of heat propagation is achieved, the long-distance heat propagation effect is captured, and the heat propagation path that the model focuses on has a clear physical meaning.

[0094] Optionally, step S303, predicting the hot spot probability of each component at the target time based on the long-range heat propagation effect characteristic representation of each component, may include the following steps:

[0095] Step S601: determining a first probability dataset and a first heat propagation dataset of hot spots for each component in a current time period, wherein the first heat propagation dataset is a data collection reflecting the characteristics of the long-range heat propagation effect of each component in the current time period;

[0096] Step S602: determining a target mapping relationship based on the first probability data set and the first heat propagation data set, wherein the target mapping relationship is used to reflect the correspondence between the hot spot probability and the long-range heat propagation effect characteristic representation;

[0097] Step S603: predicting the hot spot probability of each component at the target time based on the remote heat propagation effect characteristic representation of each component and the target mapping relationship.

[0098] In one possible embodiment, a first probability dataset and a first heat propagation dataset of hot spots for each component in the current time period are determined, wherein the first heat propagation dataset is used to reflect the data collection representing the characteristics of the long-range heat propagation effect of each component in the current time period. The original data is subjected to outlier removal and missing value interpolation. According to the hot spot probability labeling, based on historical infrared images and operation and maintenance records, a threshold method is used to automatically identify the hot spot area, and a five-level probability labeling is adopted: 0-no hot spot risk, 0.25-low risk, 0.5-medium risk, 0.75-high risk, 1-hot spot has been formed. The first probability dataset is constructed, and the spatiotemporal attention mechanism and graph convolutional network are applied to the preprocessed historical data to extract the long-range heat propagation effect characteristics of each component in each time period, including heat diffusion speed, direction, distance, etc. The feature vectors are arranged in chronological order to form the first heat propagation dataset, ensuring that the characteristics of each time point are aligned with the hot spot probability label.

[0099] In one possible embodiment, a target mapping relationship is determined based on the first probability data set and the first heat propagation data set, wherein the target mapping relationship is used to reflect the correspondence between the hot spot probability and the remote heat propagation effect feature representation. First, a multi-layer perceptron network architecture is designed. For example, the input layer receives a 128-dimensional remote heat propagation effect feature vector, including temperature gradient, electrical parameter correlation, etc. The hidden layer adopts a 3-layer fully connected structure with 256, 512, and 256 neurons respectively. Each layer is followed by an activation function, and the output layer outputs a 5-dimensional vector through a softmax function, representing the probability distribution of hot spots at five risk levels. Then, a physical constraint loss function is designed, wherein the cross entropy loss function is used to measure the difference between the predicted probability distribution and the labeled probability. The physical constraints include heat conduction continuity constraint, energy conservation constraint, and time series smoothing constraint. The heat conduction continuity constraint is used to constrain the hot spot probability change rate of adjacent components to be inversely proportional to the thermal resistance. The energy conservation constraint is used to constrain the predicted hot spot energy distribution to be consistent with the actual energy flow. The time series smoothing constraint is used to constrain the hot spot probability change of adjacent time steps to not exceed the maximum value allowed by physical laws. Then, an adaptive training strategy is adopted, the initial learning rate and early stopping mechanism are set, and K-fold cross validation is used to ensure the generalization ability of the model.

[0100] In one possible embodiment, the probability of a hot spot occurring at a target time is predicted for each component based on the long-range heat propagation effect feature representation and target mapping relationship for each component. The real-time extracted long-range heat propagation effect features are Z-score normalized. Using the mean and standard deviation calculated during the training phase, a sliding time window is constructed. The current features are concatenated with historical features to form a temporal feature sequence. This preprocessed feature is then fed into a trained multi-layer perceptron to obtain the probability distribution of hot spots at each risk level. The predicted entropy is then calculated as a measure of uncertainty. A higher entropy value indicates a lower confidence level in the model's prediction.

[0101] In a specific embodiment, the characteristic representation of the remote heat propagation effect is input into a trained multi-layer perceptron network, and the probability of occurrence of hot spots at different time and space positions is output; the training method of the multi-layer perceptron network includes: using the characteristic representation of the remote heat propagation effect obtained by the historical time data and historical space data of the substation component array, training the multi-layer perceptron network, using the probability of occurrence of hot spots at different time and space positions as labels, and adjusting the parameters of the multi-layer perceptron through the back-propagation algorithm, so that the difference between the prediction result of the multi-layer perceptron and the actual probability of occurrence of hot spots at different time and space positions is within a preset range.

[0102] Optionally, step S304, determining a hot spot probability map and a hot spot evolution direction vector based on the n hot spot probabilities of the n components, may include the following steps:

[0103] Step S701: Determine k reference hot spots based on the dynamic spatiotemporal graph;

[0104] Step S702: Determine a first capsule vector for each reference hot spot, where the first capsule vector includes a dimensional code of the corresponding reference hot spot among the k reference hot spots in the current time period, where the dimensional code includes a spatial dimensional code, a temporal dimensional code, a thermal feature dimensional code, and a dynamic evolution dimensional code;

[0105] Step S703: Determine a second capsule vector for each reference hot spot based on the first capsule vector of each reference hot spot and the long-range heat propagation effect characteristic representation of each component, wherein the second capsule vector includes a dimensional encoding of the corresponding reference hot spot among the k reference hot spots between the current moment and the target moment;

[0106] Step S704: Determine a hot spot probability map and a hot spot evolution direction vector based on the second capsule vector of each reference hot spot and the n hot spot probabilities.

[0107] In one possible embodiment, k reference hot spots are determined based on a dynamic space-time graph, where k is a positive integer. The temperature anomaly points in the dynamic space-time graph are clustered, and the spatial neighborhood radius and time window are set to form an initial set of hot spot candidates. The temperature mean, standard deviation, volume, and duration of each candidate set are calculated, and hot spots with temperatures exceeding a threshold and a duration of ≥1 hour are screened out as preliminary reference hot spots. The preliminary reference hot spots are physically verified for thermal conduction, and their thermal diffusion coefficients are calculated to determine whether they are within a reasonable range. Pseudo-hot spots that do not conform to the laws of thermodynamics are excluded. In combination with electrical parameter analysis, it is verified whether the hot spots are accompanied by current anomalies, such as local overheating caused by increased contact resistance, to enhance the reliability of hot spot identification.

[0108] In one possible embodiment, the first capsule vector of each reference hot spot is determined, wherein the first capsule vector includes the dimensional encoding of the corresponding reference hot spot among the k reference hot spots in the current time period, and the dimensional encoding includes spatial dimension encoding, temporal dimension encoding, thermal feature dimension encoding, and dynamic evolution dimension encoding. The spatial dimension encoding uses a three-dimensional vector to represent the center position of the hot spot and adds the spatial uncertainty of the Gaussian distribution. The temporal dimension encoding uses a timestamp vector to represent the time of the first appearance and duration of the hot spot, and encodes the day and night and seasonal characteristics through a periodic function. The thermal feature dimension encoding includes thermophysical parameters such as temperature peak, temperature rise rate, and thermal diffusion coefficient. The dynamic evolution dimension encoding extracts the historical evolution pattern of the hot spot through LSTM, such as linear growth and exponential growth. When initializing the vector, for the spatial dimension, it is initialized based on the cluster center position and the initial uncertainty is set to the cluster radius. For the temporal dimension, it is initialized using the timestamp of the first detection of the hot spot and the duration is set to the current value. For the thermal feature dimension, the initialization parameters are calculated based on the temperature statistics of the hot spot and the heat conduction model. For the dynamic evolution dimension, the LSTM is trained with historical data and the output hidden state is used as the initial encoding.

[0109] In one possible embodiment, the second capsule vector of each reference hot spot is determined based on the first capsule vector of each reference hot spot and the characteristic representation of the remote heat propagation effect of each component, wherein the second capsule vector includes the dimensional encoding of the corresponding reference hot spot among the k reference hot spots between the current moment and the target moment. The similarity between the remote heat propagation effect characteristics of each component and the capsule vector of each reference hot spot is calculated as the initial routing weight. The routing weight is updated iteratively so that the feature vector is more likely to be assigned to a capsule with high similarity. After each iteration, the capsule vector is updated according to the allocation result so that it more accurately represents the characteristics of the corresponding hot spot. The capsule vector of the current moment is concatenated with the capsule vector of the historical moment to form a time series input. The time series is processed using a gated recurrent unit to capture the evolution law of the hot spot from the current moment to the target moment.

[0110] In one possible embodiment, a hot spot probability map and a hot spot evolution direction vector are determined based on the second capsule vector and n hot spot probabilities of each reference hot spot. The updated capsule vector is mapped to a spatial grid through a fully connected layer. The output of each grid point represents the probability of a hot spot existing at that location. Bilinear interpolation is used to smooth the probability distribution, eliminating discontinuities at the grid boundary. Threshold segmentation is applied to extract the hot spot area, and its centroid, area and other geometric features are calculated. A differential operation is performed on the time series capsule vector to obtain the rate of change of the hot spot in each dimension. The rate of change in the spatial dimension is normalized to obtain a unit vector representing the direction of movement of the hot spot. The hot spot expansion speed is calculated by combining the thermal diffusion coefficient and the temperature gradient as the modulus of the evolution vector. The uncertainty of the probability map is estimated, the prediction variance of each grid point is calculated, and the uncertainty of the evolution direction is quantified based on the modulus of the capsule vector and the entropy of the routing weight.

[0111] In a specific embodiment, the dimensions of the capsule vector are set and the capsule vector is initialized according to the hot spots at different time and space positions; the characteristic representation of the remote heat propagation effect is input into the capsule structure, and the proportion of the remote heat propagation effect characteristic representation allocated to each capsule vector is dynamically adjusted according to the degree of coupling between the remote heat propagation effect characteristic representation and the capsule vector through a dynamic routing algorithm, and the value of the capsule vector is continuously updated; the updated capsule vector is converted into a probabilistic form through a fully connected layer and a softmax function, and a hot spot probability map containing the probability of occurrence of hot spots at different time and space positions and a vector representing the evolution direction of the hot spot is output.

[0112] In the embodiment of the present application, the dynamic routing mechanism enables the model to adaptively assign component features to the most relevant hot spot types, thereby improving prediction accuracy.

[0113] By acquiring the temporal and spatial data of the substation component array and constructing a dynamic spatiotemporal graph based on the physical characteristic constraints of the substation, the present invention can comprehensively and accurately integrate the multi-dimensional information of the substation operation process, and fully consider the interaction between temporal and spatial factors in the generation and evolution of hot spot defects. On this basis, the trained hot spot defect prediction model is used for prediction. The output hot spot probability map and the vector representing the direction of hot spot evolution can clearly show the possibility of occurrence and development trend of hot spots in different time and space positions. The cause of hot spots is further decomposed and a three-dimensional hot spot evolution sand table is constructed. It can intuitively and three-dimensionally display the dynamic changes of hot spot defects in time and space, so that operation and maintenance personnel can grasp the development trend of hot spots in advance, solving the problem that existing technologies cannot predict the dynamic changes of hot spot defects in time and space.

[0114] The present invention constructs a dynamic space-time graph through time data, space data and substation physical feature constraints, takes substation components as nodes, shadow propagation paths as edges, and calculates edge weights. It can accurately depict the complex environment in which hot spot defects in substations are generated and developed, and fully considers the interaction of various physical factors and space-time factors in the hot spot formation process, providing a solid data foundation and accurate model framework for subsequent hot spot prediction, greatly improving the accuracy and reliability of hot spot prediction.

[0115] The present invention adopts a multi-layer structure including an LSTM network layer enhanced by physical information, a graph neural network layer guided by an attention mechanism, and a probabilistic space-time capsule network layer through a spot defect prediction model. The heat conduction equation is used as a regularization constraint to guide the LSTM network to learn features that conform to physical laws and are highly correlated with hot spots; the attention mechanism is used to capture the characteristic representation of long-range heat propagation effects; the probability of occurrence of hot spots at different time and space positions is modeled through a capsule structure to obtain a hot spot probability map and a vector representing the evolution direction of the hot spots. This can effectively mine the complex patterns and features in the data and accurately predict the probability of occurrence and evolution direction of hot spots at different time and space positions.

[0116] The present invention converts the hot spot probability map into a dynamic thermal map by constructing a probability-color temperature nonlinear mapping relationship, normalizes the evolution direction vector to generate a vector thermal map, and constructs a three-dimensional substation component array sand table model in combination with the geographic coordinates, azimuth and inclination parameters of the substation components. The thermal map and the vector thermal map are integrated into the sand table model, and the multi-spectral characteristics of a real thermal imager are simulated using a ray tracing algorithm to ultimately generate a three-dimensional hot spot evolution sand table that is continuous in time and space and can be played at a variable speed. Through an intuitive and dynamic display method, operation and maintenance personnel can clearly observe the dynamic change process of hot spot defects in time and space, and have an in-depth understanding of the development trend and evolution law of hot spots, providing more intuitive and effective support for substation operation and maintenance decisions, helping to reduce the damage of hot spots to substation components and improve the operating efficiency and stability of substations.

[0117] To summarize, in an embodiment of the present application, the time data and spatial data of the substation component array in the current time period are first obtained, wherein the time data includes a first sequence corresponding to a first parameter, and the first sequence is a sequence consisting of the change of the first parameter over time in the current time period, and the first parameter includes at least one of the following: the temperature, electrical parameters and environmental data of the substation component array; the spatial data includes a second sequence corresponding to a second parameter, and the second sequence is a sequence consisting of the change of the second parameter over time in the current time period, and the second parameter includes at least one of the following: the spatial position of the hot spot and the shadow area of ​​the substation component array; then, based on the first sequence, the second sequence and the preset physical feature constraints, a dynamic spatiotemporal graph of the substation component array in the current time period is constructed, wherein the dynamic spatiotemporal graph is used to reflect the first dynamic change of the hot spot on the substation component array in the current time period; next, based on the dynamic spatiotemporal graph, the hot spot probability map and the hot spot evolution direction vector of the substation component array at the target time are predicted, wherein the target time is later than the current time; finally, based on the hot spot probability map and the hot spot evolution direction vector, the second dynamic change of the hot spot of the substation component array from the current time to the target time is predicted. Therefore, by obtaining the time data and spatial data of the substation component array in the current time period, and predicting the hot spot probability map and hot spot evolution direction vector of the substation component array at the target moment based on the dynamic time-space diagram, the second dynamic change of the hot spots of the substation component array from the current moment to the target moment can be predicted based on the hot spot probability map and the hot spot evolution direction vector, thereby being able to predict the dynamic changes of the hot spots in time and space, and then the operation and maintenance personnel can formulate effective prevention and response measures in advance based on the predicted dynamic changes of the hot spots, thereby reducing the failure risk of the substation.

[0118] The above describes in detail the method according to the embodiment of the present invention. The following provides an apparatus according to the embodiment of the present invention.

[0119] See Figure 7 , Figure 7 This is a schematic diagram of the structure of a hot spot prediction device for a substation provided in an embodiment of the present application. Figure 7As shown, the substation hot spot prediction device 800 includes an acquisition unit 801 and a processing unit 802; the acquisition unit 801 is used to acquire time data and spatial data of the substation component array in the current time period, wherein the time data includes a first sequence corresponding to a first parameter, the first sequence is a sequence composed of the change amount of the first parameter over time in the current time period, and the first parameter includes at least one of the following: temperature, electrical parameters and environmental data of the substation component array; the spatial data includes a second sequence corresponding to a second parameter, the second sequence is a sequence composed of the change amount of the second parameter over time in the current time period, and the second parameter includes at least one of the following: The spatial positions of the hot spots and shadow areas of the power station component array; a processing unit 802, for constructing a dynamic space-time diagram of the substation component array in the current time period according to the first sequence, the second sequence and the preset physical feature constraints, wherein the dynamic space-time diagram is used to reflect the first dynamic changes of the hot spots on the substation component array in the current time period; based on the dynamic space-time diagram, predicting the hot spot probability map and the hot spot evolution direction vector of the substation component array at the target moment, wherein the target moment is later than the current moment; based on the hot spot probability map and the hot spot evolution direction vector, predicting the second dynamic changes of the hot spots of the substation component array from the current moment to the target moment.

[0120] In a specific implementation, the acquisition unit 801 and the processing unit 802 in the embodiment of the present application may also execute other implementations described in the substation hot spot prediction method in the above embodiment of the present application, which will not be repeated here.

[0121] See Figure 8 , Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 8 As shown, the electronic device 900 includes a transceiver 901, a processor 902, and a memory 903, which are connected via a bus 904. The memory 903 is used to store computer programs and data, and can transmit the data stored in the memory 903 to the processor 902. The electronic device 900 can be the above-mentioned substation hot spot prediction device 800, and the processor 902 can be the above-mentioned acquisition unit 801 and processing unit 802. In the embodiment of the present application, the processor 902 is used to read the computer program in the memory 903 to execute some or all steps of the above-mentioned substation hot spot prediction method.

[0122] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. The computer program is executed by a processor to implement part or all of the steps of any substation hot spot prediction method described in the above method embodiments.

[0123] An embodiment of the present application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute part or all of the steps of any substation hot spot prediction method recorded in the above method embodiments.

[0124] It should be noted that for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all optional embodiments, and the actions and modules involved are not necessarily required by this application.

[0125] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0126] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or modules can be electrical or other forms.

[0127] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of the present embodiment according to actual needs.

[0128] In addition, the functional modules in the various embodiments of the present application may be integrated into a processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The above-mentioned integrated modules may be implemented in the form of hardware or software program modules.

[0129] If the integrated module is implemented in the form of a software program module and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a memory and includes a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned memory includes various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0130] The above is a detailed introduction to the embodiments of the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for those skilled in the art, according to the idea of ​​the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for predicting hot spots in substations, characterized in that: include: Acquiring time data and spatial data of a substation component array in a current time period, wherein the time data includes a first sequence corresponding to a first parameter, the first sequence being a sequence consisting of a change in the first parameter over time in the current time period, the first parameter including at least one of the following: temperature, electrical parameters, and environmental data of the substation component array; and the spatial data includes a second sequence corresponding to a second parameter, the second sequence being a sequence consisting of a change in the second parameter over time in the current time period, the second parameter including at least one of the following: spatial locations of hot spots and shadow areas of the substation component array; Constructing a dynamic spatiotemporal graph of the substation component array in the current time period based on the first sequence, the second sequence, and preset physical feature constraints, wherein the dynamic spatiotemporal graph is used to reflect a first dynamic change of the hot spot on the substation component array in the current time period; Predicting a hot spot probability map and a hot spot evolution direction vector of the substation component array at a target time based on the dynamic spatiotemporal graph, wherein the target time is later than the current time; A second dynamic change of the hot spots of the substation component array from the current moment to the target moment is predicted based on the hot spot probability map and the hot spot evolution direction vector.

2. The method according to claim 1, wherein The substation component array includes n components; constructing a dynamic spatiotemporal graph of the substation component array in the current time period according to the first sequence, the second sequence, and preset physical feature constraints, includes: Determining, according to the first sequence, a node characteristic of each component, wherein the node characteristic of each component is used to reflect the temperature gradient, electrical parameters, and aging index of each component in the current time period; determining, according to the second sequence, shadow propagation paths between the n components; Determining the adjacency relationship between the n components according to the shadow propagation path; Determining edge features between adjacent components among the n components based on the adjacency relationship, wherein the edge features between adjacent components are used to reflect the distance, temperature gradient, and electrical parameter correlation between the adjacent components in the current time period; Determining edge weights between the adjacent components based on the edge features between the adjacent components and the physical feature constraints; The dynamic spatiotemporal graph is constructed according to the n node features of the n components and the edge weights between the adjacent components.

3. The method according to claim 2, wherein The method of predicting the hot spot probability map and the hot spot evolution direction vector of the substation component array at the target time based on the dynamic spatiotemporal graph includes: Determine, based on the dynamic spatiotemporal graph, the time series features and spatial series features corresponding to each component, wherein the time series features are used to reflect the temporal propagation pattern of the hot spot in each component, and the spatial series features are used to reflect the spatial propagation pattern of the hot spot in each component; Determining a long-range heat propagation effect characteristic representation of each component based on the time series characteristics and spatial series characteristics corresponding to each component, wherein the long-range heat propagation effect characteristic representation of each component is used to reflect the diffusion speed, diffusion direction, and diffusion distance of the hot spot corresponding to each component; Predicting the hot spot probability of each component at the target time based on the long-range heat propagation effect characteristic representation of each component; The hot spot probability map and the hot spot evolution direction vector are determined based on the n hot spot probabilities of the n components.

4. The method according to claim 3, wherein Determining the time series characteristics and space series characteristics corresponding to each component according to the dynamic time-space graph includes: According to the dynamic spatiotemporal graph, a time series analysis is performed on the node features of each component to obtain m subsequences corresponding to each component; Perform feature extraction on the m subsequences corresponding to each component to obtain a time series feature corresponding to each component; Constructing a spatial adjacency matrix according to the edge features between the adjacent components and the edge weights between the adjacent components; According to a preset graph convolutional network, the spatial adjacency matrix and the edge features between adjacent components are processed to obtain the spatial sequence features of each component.

5. The method according to claim 3, wherein Determining the remote heat propagation effect characteristic representation of each component based on the time series characteristics and spatial series characteristics corresponding to each component includes: performing a first dot product calculation on the time series features between adjacent components according to the time series features corresponding to each component and the adjacency relationship, to obtain a first dot product calculation result between the adjacent components; performing a second dot product calculation on the spatial sequence features between adjacent components according to the spatial sequence features corresponding to each component and the adjacency relationship, to obtain a second dot product calculation result between the adjacent components; A remote heat propagation effect characteristic representation of each component is determined according to a first dot product calculation result between the adjacent components and a second dot product calculation result between the adjacent components.

6. The method according to claim 3, wherein The predicting of the hot spot probability of each component at the target time based on the remote heat propagation effect characteristic representation of each component includes: Determine a first probability dataset and a first heat propagation dataset of a hot spot for each component in the current time period, wherein the first heat propagation dataset is a data collection reflecting a characteristic representation of a long-range heat propagation effect of each component in the current time period; Determining a target mapping relationship based on the first probability dataset and the first heat propagation dataset, wherein the target mapping relationship is used to reflect a correspondence between a hot spot probability and a characteristic representation of a long-range heat propagation effect; The hot spot probability of each component at the target moment is predicted based on the remote heat propagation effect characteristic representation of each component and the target mapping relationship.

7. The method according to claim 3, wherein Determining the hot spot probability map and the hot spot evolution direction vector according to the n hot spot probabilities of the n components includes: Determining k reference hot spots according to the dynamic spatiotemporal graph; Determining a first capsule vector for each reference hot spot, wherein the first capsule vector includes a dimensional code of a corresponding reference hot spot among the k reference hot spots in the current time period, the dimensional code including a spatial dimensional code, a temporal dimensional code, a thermal feature dimensional code, and a dynamic evolution dimensional code; Determining a second capsule vector for each reference hot spot based on the first capsule vector of each reference hot spot and the characteristic representation of the long-range heat propagation effect of each component, wherein the second capsule vector includes a dimensional encoding of a corresponding reference hot spot among the k reference hot spots between the current moment and the target moment; The hot spot probability map and the hot spot evolution direction vector are determined according to the second capsule vector of each reference hot spot and the n hot spot probabilities.

8. A hot spot prediction device for a substation, characterized in that: The device includes an acquisition unit and a processing unit; The acquisition unit is configured to acquire time data and spatial data of a substation component array in a current time period, wherein the time data includes a first sequence corresponding to a first parameter, the first sequence being a sequence consisting of a change in the first parameter over time in the current time period, the first parameter including at least one of the following: temperature, electrical parameters, and environmental data of the substation component array; the spatial data includes a second sequence corresponding to a second parameter, the second sequence being a sequence consisting of a change in the second parameter over time in the current time period, the second parameter including at least one of the following: spatial locations of hot spots and shadow areas of the substation component array; The processing unit is configured to construct a dynamic spatiotemporal graph of the substation component array in the current time period based on the first sequence, the second sequence, and a preset physical feature constraint, wherein the dynamic spatiotemporal graph is configured to reflect a first dynamic change of a hot spot on the substation component array in the current time period; Predicting a hot spot probability map and a hot spot evolution direction vector of the substation component array at a target time based on the dynamic spatiotemporal graph, wherein the target time is later than the current time; A second dynamic change of the hot spots of the substation component array from the current moment to the target moment is predicted based on the hot spot probability map and the hot spot evolution direction vector.

9. An electronic device, characterized in that: include: A processor and a memory, the processor being connected to the memory, the memory being used to store a computer program, and the processor being used to execute the computer program stored in the memory, so that the electronic device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 7.

Images