Substation three-dimensional layout intelligent optimization design method based on deep learning

Abstract

The invention discloses a transformer substation three-dimensional layout intelligent optimization design method based on deep learning, and aims at transformer substation equipment space layout optimization, complex constraints are adaptively processed through a deep learning model, and a three-dimensional layout scheme considering space utilization rate, electrical safety distance and maintenance accessibility is generated. The method comprises the following steps: acquiring spatial characteristics of equipment layout through a three-dimensional layout generation module, and representing geometric distribution, electrical safety distance and topographic constraints; a data coordination module is used for coordinating a dynamic mapping relation among electrical parameters, heat distribution and layout constraints of the equipment to generate data interaction characteristics; inputting the spatial features and the data interaction features into a deep learning optimization module, and generating an optimized three-dimensional layout sequence through self-adaptive deep neural network fusion and optimization; and distributing an optimization time window according to the complexity of the layout sequence, generating a dynamic optimization adjustment strategy, and iteratively generating a final three-dimensional layout path. Intelligent optimization of the substation layout is realized through deep learning, and the space utilization rate and the operation efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of substation design and deep learning, and particularly relates to a three-dimensional layout intelligent optimization design method for a substation based on deep learning. BACKGROUND

[0002] As an important part of the power system, the equipment layout of a substation directly affects the operation efficiency, safety and maintenance convenience. Traditional three-dimensional layout design of a substation mainly relies on manual experience or rule-based computer-aided design, and needs to comprehensively consider complex factors such as electrical safety distance, terrain constraints, equipment heat distribution and maintenance accessibility. However, the existing methods have the following shortcomings: first, manual design is low in efficiency and difficult to cope with complex terrain and multiple constraint scenarios; second, traditional optimization algorithms lack adaptability and are difficult to dynamically balance space utilization and electrical safety requirements; third, the existing methods are insufficient in dynamic interaction processing of device parameters and layout constraints, resulting in poor adaptability of the layout scheme in actual operation. In recent years, deep learning technology has shown great potential in complex optimization problems, but its application in three-dimensional layout optimization of a substation is still in the exploratory stage, and there is a lack of adaptive optimization methods for substation-specific constraints. Therefore, an intelligent optimization design method based on deep learning is urgently needed to realize efficient and accurate three-dimensional layout optimization of a substation. SUMMARY

[0003] The present application aims to provide a three-dimensional layout intelligent optimization design method for a substation based on deep learning, which addresses the problems of low manual efficiency, lack of adaptability of optimization algorithms and insufficient dynamic interaction processing in traditional substation layout design. By using an adaptive deep neural network to process complex constraints of a substation, the method generates a three-dimensional layout scheme that takes into account space utilization, electrical safety distance and maintenance accessibility, thereby improving the intelligent level and operation efficiency of layout design.

[0004] Specifically, the present application provides a three-dimensional layout intelligent optimization design method for a substation based on deep learning, which is applied to the spatial layout optimization of substation equipment. The method uses a deep learning model to adaptively process complex constraints of a substation, and generates a three-dimensional layout scheme that takes into account space utilization, electrical safety distance and maintenance accessibility. Specifically, the method includes the following steps:

[0005] According to the received substation design requirements, the spatial characteristics of the equipment layout are collected by a three-dimensional layout generation module, wherein the spatial characteristics represent the geometric distribution, electrical safety distance and terrain constraints of the equipment in three-dimensional space; the data interaction characteristics between the substation equipment data and the three-dimensional layout generation module are coordinated by a data coordination module, wherein the data interaction characteristics represent the dynamic mapping relationship between the electrical parameters, thermal distribution and layout constraints of the equipment; the spatial characteristics and the data interaction characteristics are input into a deep learning optimization module, which fuses the spatial characteristics and data interaction characteristics through an adaptive deep neural network to generate an optimized three-dimensional layout scheme, which meets the space utilization and maintenance accessibility requirements of the substation.

[0006] Further, the method further comprises: performing deep learning processing on the feature extraction module of the spatial characteristic change to generate a first layout feature set, wherein the output of the first layout feature set represents the spatial characteristic set of the three-dimensional layout generation module, including the device spacing and terrain adaptability; performing deep learning processing on the feature extraction module of the data interaction characteristic change to generate a second interaction feature set, wherein the output of the second interaction feature set represents the interaction characteristic set of the data coordination module, including the dynamic association of electrical parameters and thermal distribution; constructing a substation three-dimensional layout optimization deep model according to the first layout feature set and the second interaction feature set, which optimizes the substation layout by adaptively fusing the spatial characteristics and data interaction characteristics.

[0007] Further, the step of collecting the spatial characteristics of the equipment layout by the three-dimensional layout generation module comprises: determining the layout change data of the first layout feature set generated according to the design requirements, wherein the layout change data includes the geometric vector of the spatial distribution of the equipment, the weight of the electrical safety distance and the terrain constraint factor; calculating the feature intensity of each three-dimensional layout generation module according to the layout change data through a convolutional neural network, wherein the feature intensity reflects the spatial adaptability of the equipment layout; generating the spatial characteristics according to the feature intensity.

[0008] Further, the step of coordinating the data interaction characteristics between the substation equipment data and the three-dimensional layout generation module by the data coordination module comprises: determining the data interaction change data of the second interaction feature set generated according to the design requirements, wherein the data interaction change data includes the transmission amplitude of the electrical parameters of the equipment, the weight of the thermal distribution and the data flow factor; converting the transmission amplitude, the weight of the thermal distribution and the flow factor of each data coordination module into the data interaction characteristics through a recurrent neural network.

[0009] Further, the step of fusing the spatial features and the data interaction features by the adaptive deep neural network comprises: acquiring the spatial features sent by the three-dimensional layout generation module and the data interaction features sent by the data coordination module; fusing the spatial features and the data interaction features by the adaptive deep neural network to generate an optimized three-dimensional layout scheme, which optimizes the layout by dynamically adjusting the electrical safety distance and the maintenance channel width.

[0010] Further, the step of fusing the spatial features and the data interaction features by the adaptive deep neural network comprises: inputting the spatial features and the data interaction features into a multi-layer perceptron to generate a fusion feature vector, which represents the comprehensive constraints of the substation layout; processing the fusion feature vector by a residual neural network and an attention mechanism to generate an optimized three-dimensional layout sequence, which prioritizes the electrical safety and the maintenance accessibility.

[0011] Further, the step of generating the optimized three-dimensional layout scheme comprises: allocating an optimization time window according to the complexity of the three-dimensional layout sequence to generate a dynamic optimization adjustment strategy, which adaptively adjusts the wiring efficiency and the space utilization in the substation layout; inputting the dynamic optimization adjustment strategy into a layout control module to generate an optimized three-dimensional layout path through iterative optimization.

[0012] Further, the step of allocating the optimization time window according to the complexity of the three-dimensional layout sequence comprises: performing feature extraction on the three-dimensional layout sequence by a multi-scale convolutional neural network to obtain first-scale features S1 and second-scale features S2, which respectively represent the local electrical constraints and the global spatial complexity of the three-dimensional layout sequence; determining the complexity of the three-dimensional layout sequence according to the formula calculating a relative importance distribution, where R represents the relative importance distribution, S1 and S2 represent the complexity of the first and second scale features respectively, w1 and w2 are weight coefficients, and ∈ is a small constant to prevent division by zero; determining the optimization time window allocation method according to the relative importance distribution to generate the dynamic optimization adjustment strategy.

[0013] Further, the step of determining the optimization time window allocation method according to the relative importance distribution comprises: generating a dynamic threshold T corresponding to the relative importance distribution by an adaptive filter, where T = σ(R), and σ is a Sigmoid function used to normalize the relative importance distribution; calculating a distribution factor F according to the relative importance distribution and the dynamic threshold, where F represents a distribution factor; the optimization time window distribution mode is determined according to the distribution factor, wherein a longer time window is distributed to optimize the complex layout when F is greater than a preset threshold, otherwise a shorter time window is distributed to improve the calculation efficiency, and the dynamic optimization adjustment strategy is generated.

[0014] Further, the step of inputting the dynamic optimization adjustment strategy into a layout control module to generate an optimized three-dimensional layout path through iterative optimization comprises: converting the dynamic optimization adjustment strategy into a control vector, and superimposing the timing information and substation operation constraints of the three-dimensional layout sequence; processing the control vector through a multi-layer residual network and an attention mechanism to generate a preliminary layout path, which gives priority to electrical safety distance and maintenance channel accessibility; and generating a final substation three-dimensional layout path through iterative optimization of the preliminary layout path through a feedback network.

[0015] The application overcomes the limitations of low efficiency and difficulty in dealing with complex terrain and multi-constraint scenarios of artificial design by adaptively processing the complex constraints of substations through a deep learning model; solves the problem of lack of dynamic balance between space utilization and electrical safety requirements of traditional optimization algorithms by adaptively fusing spatial features and data interaction features through an adaptive deep neural network, realizes adaptive optimization of the layout scheme; enhances the dynamic interaction processing capability of device parameters and layout constraints by dynamically adjusting the electrical safety distance and the width of the maintenance channel, and improves the adaptability of the layout scheme in actual operation; and the finally generated optimized three-dimensional layout scheme significantly improves the space utilization, wiring efficiency and maintenance accessibility of the substation, providing technical support for efficient operation of the power system. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A substation three-dimensional layout intelligent optimization design method flowchart is provided for the application;

[0017] Figure 2 A method flowchart for constructing a substation three-dimensional layout optimization deep model is provided for the application;

[0018] Figure 3 A method flowchart for fusing the spatial features and data interaction features by the deep learning optimization module is provided for the application. DETAILED DESCRIPTION

[0019] To make the purpose, technical scheme and advantages of the application clearer, the technical scheme in the application will be described clearly and completely below in conjunction with the drawings in the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the application.

[0020] The terms "first", "second", etc. are used only for the purpose of description and should not be understood as indicating or implying relative importance or implying the number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features; in the description of the present application, unless otherwise stated, the meaning of "multiple" is two or more.

[0021] In order to more clearly illustrate the technical solutions of the present application, the present application will be described in detail below in combination with specific embodiments, but should not be understood as limiting the scope of protection of the present application.

[0022] The present embodiment is directed to a substation three-dimensional layout intelligent optimization design method based on deep learning, which provides a specific implementation mode, as shown in Figure 1 focuses on adaptive processing of substation complex constraints through a deep learning model to generate a three-dimensional layout scheme that takes into account space utilization, electrical safety distance, and maintenance accessibility.

[0023] Step S01: According to the received substation design requirements, the spatial features of the equipment layout are collected by the three-dimensional layout generation module, wherein the spatial features represent the geometric distribution of the equipment in the three-dimensional space, the electrical safety distance and the terrain constraints.

[0024] Specifically, the substation design requirements are imported through professional design software (such as AutoCAD or Revit), including equipment list (such as transformers, circuit breakers), size specifications and site terrain data.

[0025] Further, the three-dimensional layout generation module uses laser radar scanning equipment and geographic information system (GIS) to generate a three-dimensional model of the site and obtain geometric distribution data such as the position coordinates, height and footprint area of the equipment.

[0026] It can be understood that in order to meet the electrical safety distance requirements, the module calculates the equipment spacing according to the national grid standard (such as GB 50053-94), for example, the spacing between high-voltage circuit breakers is not less than 2.5 meters. At the same time, the terrain data (such as slope, ground bearing capacity) is analyzed through point cloud processing technology to extract the terrain constraint features. These spatial features are stored in the form of high-dimensional vectors, representing the geometric distribution of the equipment in the three-dimensional space, the electrical safety distance and the terrain constraints, ensuring that the layout adapts to the site conditions.

[0027] Step S02: The data interaction features between the substation equipment data and the three-dimensional layout generation module are coordinated by the data coordination module, wherein the data interaction features represent the dynamic mapping relationship between the electrical parameters, thermal distribution and layout constraints of the equipment.

[0028] Specifically, the data coordination module extracts electrical parameters (e.g., rated power, current load of 110 kV transformer) and operational data (e.g., thermal distribution characteristics during device operation) from the substation equipment management system.

[0029] Further, the module interacts with the three-dimensional layout generation module through the data interface, analyzes the dynamic relationship between electrical parameters and layout constraints, and evaluates the impact of device thermal distribution on safety distance, for example. Time series analysis techniques can be used to handle dynamic changes in device operating conditions, such as heat accumulation in transformers under high load, to generate data interaction feature vectors.

[0030] Understandably, these feature vectors represent the dynamic mapping relationship between device electrical parameters, thermal distribution, and layout constraints, ensuring that the layout scheme adapts to device operating requirements.

[0031] Step S03: Input the spatial features and data interaction features into the deep learning optimization module. The deep learning optimization module fuses the spatial features and data interaction features through an adaptive deep neural network to generate an optimized three-dimensional layout scheme that meets the substation's space utilization and maintenance accessibility requirements.

[0032] Specifically, the spatial features and data interaction features are input into the deep learning optimization module in the form of high-dimensional vectors. The module is based on a pre-trained deep neural network and uses a combination of convolutional neural networks (CNN) and gated recurrent units (GRU).

[0033] Further, CNN processes spatial features to extract spatial relationships between geometric distribution and terrain constraints between devices; GRU analyzes data interaction features to predict the dynamic impact of thermal distribution on layout.

[0034] Understandably, the deep learning optimization module adjusts network weights through adaptive optimization algorithms (such as RMSprop) to dynamically balance space utilization, electrical safety distance, and maintenance accessibility. For example, the module ensures that the maintenance channel width is not less than 1.5 meters while optimizing the site utilization rate to more than 80%. Finally, the module outputs an optimized three-dimensional layout scheme in the form of a BIM (Building Information Modeling) file, meeting the substation's space utilization and maintenance accessibility requirements.

[0035] The embodiment adaptively processes complex constraints of a transformer substation through a deep learning model, generates a three-dimensional layout scheme considering space utilization, electrical safety distance and maintenance accessibility, and overcomes the limitations of traditional manual design, such as low efficiency, difficulty in dealing with complex terrain and multiple constraint scenarios; the three-dimensional layout generation module collects device space features, and the data coordination module processes the dynamic mapping relationship between electrical parameters and thermal distribution of the device, solving the deficiency of traditional methods in dynamic interaction processing of device parameters and layout constraints; the adaptive deep neural network fuses space features and data interaction features, realizes intelligent optimization of the layout scheme, significantly improves the space utilization, wiring efficiency and maintenance accessibility of the transformer substation, and ensures that the layout scheme meets the actual operation requirements, providing technical support for efficient operation of the power system.

[0036] In some embodiments, as shown in Figure 2 For the deep learning-based intelligent optimization design method of transformer substation three-dimensional layout, a specific implementation is provided, focusing on constructing a three-dimensional layout optimization deep model of the transformer substation, and optimizing the transformer substation layout by adaptively fusing space features and data interaction features.

[0037] Step S21: The feature extraction module for space feature changes is processed by deep learning to generate a first layout feature set, wherein the output of the first layout feature set represents a space feature set of the three-dimensional layout generation module, including device spacing and terrain adaptability.

[0038] Specifically, the feature extraction module receives the space features collected from the three-dimensional layout generation module, which represent the geometric distribution of the device in the three-dimensional space, the electrical safety distance and the terrain constraints. For example, the module processes the position coordinates, spacing data and site slope information of the high-voltage circuit breaker and transformer in the transformer substation.

[0039] Further, the feature extraction module uses a convolutional neural network (CNN) to process the space features by deep learning, extracts device spacing features (such as circuit breaker spacing 2.5 meters) and terrain adaptability features (such as stability when the slope is less than 15 degrees). It can be understood that the CNN generates a first layout feature set by multiple convolution operations, outputs a high-dimensional vector set, represents device spacing and terrain adaptability, and ensures that the space feature set reflects the geometric and environmental constraints of the transformer substation layout.

[0040] Step S22: The feature extraction module for data interaction feature changes is processed by deep learning to generate a second interaction feature set, wherein the output of the second interaction feature set represents an interaction feature set of the data coordination module, including the dynamic association between electrical parameters and thermal distribution.

[0041] Specifically, the feature extraction module receives data interaction features generated from the data coordination module, which represent the dynamic mapping relationship between electrical parameters, thermal distribution and layout constraints. For example, the module processes the rated power, current load and thermal distribution data of a 110 kV transformer during operation.

[0042] Further, the feature extraction module uses a recurrent neural network (RNN) to analyze the dynamic changes of the data interaction features, such as the impact of heat accumulation on the layout safety distance when the transformer is under high load. Through the time series processing capability of the RNN, a second set of interaction features can be generated, outputting a set of vectors representing the dynamic correlation between electrical parameters (such as voltage stability) and thermal distribution (such as heat dissipation requirements).

[0043] As can be understood, these vector sets reflect the interaction between device operating characteristics and layout constraints, ensuring that the layout adapts to the operating requirements of the substation.

[0044] Step S23: constructing a three-dimensional layout optimization deep model of the substation based on the first set of layout features and the second set of interaction features, the three-dimensional layout optimization deep model of the substation optimizing the layout of the substation by adaptively fusing the spatial features and data interaction features.

[0045] Specifically, the first set of layout features and the second set of interaction features are input into the three-dimensional layout optimization deep model of the substation in the form of high-dimensional vectors, and the model is constructed based on a deep learning framework (such as TensorFlow).

[0046] Further, the model uses a multi-layer perceptron (MLP) to fuse the first set of layout features (representing device spacing and terrain adaptability) and the second set of interaction features (representing the dynamic correlation between electrical parameters and thermal distribution), generating a comprehensive feature representation. The model weights can be adjusted by an adaptive optimization algorithm (such as the Adam optimizer) to dynamically balance spatial utilization and electrical safety requirements. For example, the model optimizes device layout to ensure that the maintenance passage width is not less than 1.5 meters, while maximizing site utilization. As can be understood, the three-dimensional layout optimization deep model of the substation outputs an optimized three-dimensional layout scheme in a BIM format file, meeting the spatial utilization and maintenance accessibility requirements of the substation.

[0047] In some embodiments, for the deep learning-based intelligent optimization design method of the three-dimensional layout of the substation, a specific implementation is provided, focusing on collecting spatial features of device layout by the three-dimensional layout generation module, generating spatial features representing the geometric distribution of devices in three-dimensional space, electrical safety distance and terrain constraints.

[0048] Specifically, the first set of layout features is determined according to the layout change data generated by the design requirements, wherein the layout change data includes geometric vectors of device spatial distribution, electrical safety distance weights and terrain constraint factors.

[0049] It can be understood that the first set of layout features is generated by a feature extraction module based on the spatial features collected by the three-dimensional layout generation module, representing the geometric distribution, electrical safety distance, and terrain constraints of the equipment in three-dimensional space. Design requirements are imported through substation design software (such as Bentley OpenUtilities), including equipment types (such as high-voltage circuit breakers, transformers), sizes, and site terrain data.

[0050] Layout change data is generated by analyzing the three-dimensional coordinates, sizes, and site terrain of the equipment. For example, for a 110kV transformer, the module extracts its position coordinates (x, y, z), footprint area (such as 5m x 4m) as a geometric vector, calculates the electrical safety distance weight (such as the distance between high-voltage devices is not less than 2.5 meters) according to the national grid standard (such as GB 50053-94), and extracts the terrain constraint factor (such as the stability coefficient of slope less than 10 degrees) through terrain analysis.

[0051] It can be understood that these data are stored in vector form, constituting layout change data, reflecting the spatial characteristics of equipment layout.

[0052] Further, according to the layout change data, the feature strength of each three-dimensional layout generation module is calculated by a convolutional neural network, which reflects the spatial adaptability of the equipment layout.

[0053] Specifically, the layout change data (including geometric vectors, electrical safety distance weights, and terrain constraint factors) are input into a convolutional neural network (CNN) to calculate the feature strength of each three-dimensional layout generation module.

[0054] The CNN uses a multi-layer convolutional structure (such as 3 layers of convolutional layers, each with 32 filters) to extract features from the geometric vector, analyze the spatial relationship between devices, such as the relative distance between the circuit breaker and the transformer. The electrical safety distance weight is used to weight the convolution output to ensure the priority of the safety distance, and the terrain constraint factor is integrated into the feature calculation through the pooling operation to reflect the site adaptability. For example, the CNN can identify areas with a large slope and reduce the layout priority of that area.

[0055] It can be understood that the feature strength is output in numerical form (such as a normalized value from 0 to 1), representing the pros and cons of the equipment layout in spatial adaptability, providing a basis for subsequent spatial feature generation.

[0056] Further, the spatial features are generated according to the feature strength.

[0057] Specifically, the feature strength is processed to generate spatial features, representing the geometric distribution, electrical safety distance, and terrain constraints of the equipment in three-dimensional space.

[0058] The feature intensity vector is mapped to a high-dimensional spatial feature vector by a fully connected layer, for example, converting the feature intensity of a circuit breaker (reflecting its safety distance from adjacent equipment and terrain adaptability) to a 128-dimensional vector.

[0059] It can be understood that the generated spatial features are checked in combination with design requirements, for example, by comparing the electrical safety distance weight with the specification requirement, to ensure that the layout meets the safety standard. Finally, the spatial features are stored in the form of a vector set, which is used as input to the deep learning optimization module to optimize the three-dimensional layout of the substation.

[0060] In some embodiments, for the deep learning-based intelligent optimization design method of substation three-dimensional layout, a specific implementation is provided, focusing on coordinating the data interaction features between the substation equipment data and the three-dimensional layout generation module through the data coordination module, generating data interaction features representing the dynamic mapping relationship between the electrical parameters of the equipment, the heat distribution and the layout constraints.

[0061] Specifically, the second interaction feature set is determined to generate data interaction change data according to design requirements, wherein the data interaction change data includes transmission amplitude of electrical parameters of equipment, heat distribution weight and data flow direction factor.

[0062] It can be understood that the second interaction feature set is generated by the feature extraction module based on the data interaction features between the substation equipment data and the three-dimensional layout generation module coordinated by the data coordination module, which represent the dynamic mapping relationship between the electrical parameters of the equipment, the heat distribution and the layout constraints. The design requirements are imported through the substation management system, including the electrical parameters of the equipment (such as the rated current and voltage level of a 220kV circuit breaker) and the operating data (such as the heat distribution characteristics of a transformer during operation). Further, the data interaction change data is generated by analyzing the dynamic changes of the electrical parameters of the equipment. For example, for a transformer, the module extracts the transmission amplitude of its current load (such as peak current 1000A), the heat distribution weight (such as the heat dissipation demand coefficient of the high temperature area 0.8) and the data flow direction factor (such as the transmission priority of the parameters from the equipment database to the layout module). It can be understood that these data are stored in the form of vectors to constitute the data interaction change data, reflecting the dynamic interaction between the electrical parameters and the layout constraints.

[0063] Further, the transmission amplitude, heat distribution weight and data flow direction factor of each data coordination module are converted into the data interaction features by a recurrent neural network.

[0064] Specifically, the data interaction variation data (including transmission amplitude, thermal distribution weight, and data flow direction factor) are input into a recurrent neural network (RNN) to generate data interaction features. Further, the RNN adopts a long short-term memory network (LSTM) structure, containing 2 layers of LSTM units, each layer having 128 neurons, to process the time sequence characteristics of the transmission amplitude, thermal distribution weight, and data flow direction factor. For example, the LSTM analyzes the changes in the thermal distribution weight of the transformer under different loads to predict the influence of high temperature on the layout safety distance. The RNN can be trained by a gradient descent optimization algorithm (such as RMSprop) to enhance its modeling capability for dynamic data. Understandably, the RNN outputs a data interaction feature vector, for example, a 128-dimensional vector, representing the dynamic mapping relationship between the electrical parameters, thermal distribution, and layout constraints of the equipment. These feature vectors are used for subsequent deep learning optimization to ensure that the layout scheme adapts to the operation requirements of the substation.

[0065] In some embodiments, as shown in Figure 3 For the deep learning-based intelligent optimization design method for three-dimensional layout of substations, a specific implementation is provided, focusing on using an adaptive deep neural network to fuse spatial features and data interaction features by a deep learning optimization module to generate an optimized three-dimensional layout scheme that meets the space utilization and maintenance accessibility requirements of the substation.

[0066] Step S51: Obtain the spatial features sent by the three-dimensional layout generation module in real time and the data interaction features sent by the data coordination module.

[0067] Specifically, the deep learning optimization module receives spatial features from the three-dimensional layout generation module, representing the geometric distribution of equipment in three-dimensional space, electrical safety distance, and terrain constraints, and receives data interaction features from the data coordination module, representing the dynamic mapping relationship between electrical parameters, thermal distribution, and layout constraints of the equipment.

[0068] Further, the spatial features are transmitted from the three-dimensional layout generation module through a real-time data interface (such as RESTAPI), containing the position coordinates of the equipment (such as the x, y, and z coordinates of the transformer), safety distance data (such as the distance between high-voltage equipment of 2.5 meters), and terrain features (such as slope constraints). The data interaction features are obtained through a database query interface, for example, the current load and thermal distribution data of a 220kV circuit breaker. The features can be normalized by a data preprocessing pipeline to ensure consistent feature vector formats (such as unified 128-dimensional vectors).

[0069] Understandably, these features are input into the deep learning optimization module in the form of real-time streams to ensure data freshness and the dynamics of layout optimization.

[0070] Step S52: generating an optimized three-dimensional layout scheme by fusing the spatial features and the data interaction features through an adaptive deep neural network, the scheme optimizing the layout by dynamically adjusting the electrical safety distance and the maintenance passage width.

[0071] Specifically, the adaptive deep neural network adopts a combined architecture of convolutional neural network (CNN) and long short-term memory network (LSTM) and runs in a deep learning framework (such as PyTorch).

[0072] Further, the CNN processes the spatial features, extracting the geometric relationship and terrain constraint features between devices; the LSTM analyzes the time sequence change of the data interaction features, predicting the influence of the heat distribution on the layout. For example, the network dynamically adjusts the distance of the transformer from the adjacent devices according to the heat distribution data of the transformer, ensuring that the heat dissipation requirement meets the safety requirement. The network weight can be adjusted through an adaptive optimization algorithm (such as Adam optimizer) to balance the space utilization and the maintenance accessibility, for example, to optimize the maintenance passage width to more than 1.5 meters while ensuring that the site utilization rate reaches 85%.

[0073] Understandably, the network outputs an optimized three-dimensional layout scheme, which is exported in BIM (Building Information Modeling) format, containing device coordinates, distances, and passage layout, meeting the requirements of electrical safety distance and maintenance accessibility of the substation.

[0074] In some embodiments, for the deep learning-based intelligent optimization design method of substation three-dimensional layout, a specific implementation is provided, focusing on generating an optimized three-dimensional layout sequence by fusing spatial features and data interaction features through an adaptive deep neural network, and prioritizing the requirements of electrical safety and maintenance accessibility.

[0075] Specifically, the spatial features and the data interaction features are input into a multi-layer perceptron to generate a fusion feature vector, which represents the comprehensive constraints of the substation layout.

[0076] It can be understood that the spatial features (characterizing the geometric distribution of the equipment in three-dimensional space, the electrical safety distance, and the terrain constraints) and the data interaction features (characterizing the dynamic mapping relationship of the electrical parameters, the thermal distribution, and the layout constraints of the equipment) are input into the multi-layer perceptron (MLP) through the real-time data interface. Further, the MLP adopts a three-layer fully connected neural network structure, with 256 neurons in each layer, to process the spatial features (such as the 3-meter safety distance between the transformer and the circuit breaker, the terrain slope data) and the data interaction features (such as the current load and thermal distribution of the 220kV circuit breaker). The feature vectors can be normalized to a uniform dimension (such as 128 dimensions) through standardized preprocessing, ensuring input consistency. It can be understood that the MLP fuses the two types of features through matrix operations to generate a fused feature vector, which represents the comprehensive constraints of the substation layout, such as the influence of equipment spacing on thermal distribution or the limitation of terrain on safety distance, reflecting the overall optimization requirements of the layout.

[0077] Further, the fused feature vector is processed through a residual neural network and an attention mechanism to generate an optimized three-dimensional layout sequence that prioritizes electrical safety and maintenance accessibility.

[0078] Specifically, the fused feature vector is input into a combined model of a residual neural network (ResNet) and an attention mechanism, which runs on a deep learning framework (such as TensorFlow). Further, the ResNet adopts an 18-layer residual structure to process the fused feature vector and optimize complex constraint relationships, such as preserving the detailed features of equipment spacing and terrain constraints through residual connections. The attention mechanism (based on a self-attention model) assigns weights to prioritize electrical safety distances (such as a high-voltage equipment spacing not less than 2.5 meters) and maintenance channel widths (such as not less than 1.5 meters). For example, the attention mechanism enhances the weight of the thermal distribution features to dynamically adjust the layout of high-temperature equipment to meet the heat dissipation requirements. It can be understood that the model outputs an optimized three-dimensional layout sequence, which includes equipment coordinates, spacing, and channel layout, stored in JSON format, prioritizing electrical safety and maintenance accessibility requirements to ensure that the layout scheme is applicable to the actual operation of the substation.

[0079] In some embodiments, for the deep learning-based intelligent optimization design method of substation three-dimensional layout, a specific implementation is provided, focusing on generating an optimized three-dimensional layout scheme, by allocating an optimization time window according to the complexity of the three-dimensional layout sequence and generating a dynamic optimization adjustment strategy to iteratively generate a three-dimensional layout path that meets the space utilization and wiring efficiency of the substation.

[0080] Specifically, an optimization time window is allocated according to the complexity of the three-dimensional layout sequence to generate a dynamic optimization adjustment strategy that adaptively adjusts the wiring efficiency and space utilization in the substation layout.

[0081] It can be understood that the three-dimensional layout sequence is generated by the deep learning optimization module based on fused spatial features (representing the geometric distribution of devices in three-dimensional space, electrical safety distance, and terrain constraints) and data interaction features (representing the dynamic mapping relationship between device electrical parameters, thermal distribution, and layout constraints), containing device coordinates, spacing, and preliminary layout information. Further, by analyzing the complexity of the three-dimensional layout sequence (such as the number of devices, the number of constraint conditions, and the degree of terrain change), a complexity-based scheduling algorithm is used to allocate an optimization time window. For example, for a complex substation layout containing 50 devices, a longer time window (such as 500 ms) is allocated to handle high-complexity areas; for a simple layout, a shorter time window (such as 100 ms) is allocated. It can be understood that the dynamic optimization adjustment strategy adjusts the wiring efficiency (such as optimizing cable path length) and space utilization (such as site utilization of 85%) through priority sorting, generates a strategy file containing time window allocation and optimization priority, and is used to guide subsequent iterative optimization.

[0082] Further, the dynamic optimization adjustment strategy is input into the layout control module to generate an optimized three-dimensional layout path through iterative optimization.

[0083] Specifically, the dynamic optimization adjustment strategy is input into the layout control module in JSON format, which runs on a computing platform (such as a high-performance server). Further, the layout control module uses an iterative optimization algorithm (such as a genetic algorithm) to adjust the three-dimensional layout sequence according to the dynamic optimization adjustment strategy. For example, the module optimizes the spacing between high-voltage devices (not less than 2.5 meters) and the width of the maintenance channel (not less than 1.5 meters) according to the strategy, while adjusting the device coordinates through multiple iterations (for example, 10 times) to maximize space utilization. It can be understood that the iterative optimization process combines substation operation constraints (such as minimizing cable routing length) to generate an optimized three-dimensional layout path, output as a BIM (Building Information Modeling) format file, containing the final coordinates of the devices, the routing path, and the layout of the maintenance channel, meeting the space utilization and routing efficiency requirements of the substation.

[0084] In some embodiments, for the deep learning-based intelligent optimization design method for substation three-dimensional layout, a specific implementation is provided, focusing on allocating an optimization time window according to the complexity of the three-dimensional layout sequence, and generating a dynamic optimization adjustment strategy to adaptively adjust the wiring efficiency and space utilization in the substation layout.

[0085] Specifically, the three-dimensional layout sequence is feature-extracted by a multi-scale convolutional neural network to obtain first-scale features S1 and second-scale features S2, representing local electrical constraints and global spatial complexity of the three-dimensional layout sequence, respectively.

[0086] It can be understood that the three-dimensional layout sequence is generated by the deep learning optimization module based on fused spatial features (characterizing the geometric distribution of devices in three-dimensional space, electrical safety distance, and terrain constraints) and data interaction features (characterizing the dynamic mapping relationship between device electrical parameters, thermal distribution, and layout constraints), and contains device coordinates, spacing, and preliminary layout information.

[0087] A multi-scale convolutional neural network (CNN) uses two-scale convolution kernels (such as 3x3 and 5x5) to process the three-dimensional layout sequence, extracts first-scale features S1 representing local electrical constraints, such as the 2.5-meter safety spacing requirement between high-voltage circuit breakers and transformers, and extracts second-scale features S2 representing global spatial complexity, such as the utilization rate of the entire substation site and the degree of terrain changes. The feature extraction process can be stabilized by batch normalization technology to generate S1 and S2 vectors (such as 64-dimensional vectors).

[0088] It can be understood that these features reflect the local and global constraint characteristics of the layout sequence, providing a basis for optimization time window allocation.

[0089] Further, the relative importance distribution is calculated according to the formula where R represents the relative importance distribution, S1 and S2 represent the complexity of the first and second scale features, w1 and w2 are weight coefficients, and ∈ is a small constant to prevent division by zero.

[0090] Specifically, the formula is used to calculate the relative importance distribution, which measures the relative importance of local electrical constraints and global spatial complexity.

[0091] It can be understood that S1 and S2 are extracted from the multi-scale CNN, and the weight coefficients w1 and w2 are set according to the design requirements of the substation, such as for a layout with high-voltage device density, setting w1 = 0.7 (emphasizing local electrical constraints) and w2 = 0.3 (global spatial complexity); ∈ is set to 0.001 to prevent division by zero. The formula calculation can be realized through a computing platform (such as the NumPy library of Python), and the relative importance distribution R is generated in the form of a numerical vector (range 0 to 1), reflecting the optimization priority of different regions. For example, the R value of a high electrical constraint region (such as near a transformer) is higher, and optimization resources are preferentially allocated.

[0092] It can be understood that the R vector guides the allocation of the optimization time window, improving the computational efficiency of optimizing complex layouts.

[0093] Further, the optimization time window allocation method is determined according to the relative importance distribution, and the dynamic optimization adjustment strategy is generated.

[0094] In particular, the relative importance distribution R is used to determine the allocation of optimization time windows, generating a dynamic optimization adjustment strategy. Further, the time windows are allocated by a threshold comparison method: for regions with a high R value (e.g., R > 0.6, corresponding to high electrical constraint regions), a longer time window (e.g., 600 ms) is allocated to ensure accurate optimization; for regions with a low R value (e.g., R < 0.3, corresponding to simple terrain regions), a shorter time window (e.g., 100 ms) is allocated to improve efficiency. The allocation of time windows can be achieved by a scheduling algorithm (e.g., priority queue), generating a dynamic optimization adjustment strategy, output as a JSON format file containing time window allocation and optimization priority.

[0095] It can be understood that this strategy adaptively adjusts the wiring efficiency (e.g., optimizes the cable path length) and space utilization (e.g., site utilization rate of 85%) in the substation layout, providing guidance for the iterative optimization of the layout control module.

[0096] In some embodiments, for a deep learning-based intelligent optimization design method for three-dimensional substation layout, a specific implementation is provided, focusing on determining the allocation of optimization time windows according to the relative importance distribution, generating a dynamic optimization adjustment strategy to adaptively adjust the wiring efficiency and space utilization in the substation layout.

[0097] In particular, a dynamic threshold T corresponding to the relative importance distribution is generated by an adaptive filter, where T = σ(R), σ is a Sigmoid function, used to normalize the relative importance distribution.

[0098] It can be understood that the relative importance distribution R is calculated by the above formula , representing the relative importance of local electrical constraints and global spatial complexity of the three-dimensional layout sequence.

[0099] Further, the adaptive filter uses a Sigmoid function to normalize R, generating a dynamic threshold T ranging from 0 to 1. For example, for a high-voltage equipment-intensive area in the substation, R = 0.8, the Sigmoid function calculates T ≈ 0.69, reflecting the high optimization priority of this area. This calculation can be achieved by the Sigmoid module of a deep learning framework (e.g., PyTorch), ensuring that the threshold dynamically adapts to the complexity of the layout.

[0100] It can be understood that the dynamic threshold T is used for subsequent allocation factor calculation, guiding the allocation of optimization time windows.

[0101] Further, an allocation factor F is calculated according to the relative importance distribution and the dynamic threshold, where

[0102] F represents the allocation factor.

[0103] Specifically, the allocation factor F is calculated by the formula , where R is the relative importance distribution, T is the dynamic threshold, and ∈ is a small constant (set to 0.001) to prevent division by zero.

[0104] In substation layout optimization, R and T are obtained from the previous step, for example, for a high electrical constraint area near the transformer (R = 0.8, T = 0.69), F ≈ 1.16 is calculated. The formula can be implemented by the NumPy library in Python to generate the allocation factor F, which is stored in numerical form, reflecting the optimization resource requirements of different layout areas.

[0105] It can be understood that the higher the F value, the higher the complexity of the area, and more optimization resources need to be allocated to ensure that the layout meets the electrical safety distance and space utilization requirements.

[0106] Further, the allocation factor is used to determine the optimization time window allocation method, wherein a longer time window is allocated to optimize complex layouts when F is greater than a preset threshold, otherwise a shorter time window is allocated to improve computational efficiency, and the dynamic optimization adjustment strategy is generated.

[0107] Specifically, the allocation factor F is used to determine the optimization time window allocation method, and the preset threshold is set to 1.0.

[0108] For areas with F > 1.0 (such as high-voltage equipment intensive areas, F = 1.16), a longer time window (such as 800 ms) is allocated to accurately optimize complex layouts, such as adjusting the safety distance between the transformer and the circuit breaker (not less than 2.5 meters); for areas with F < 1.0 (such as low-complexity terrain areas, F = 0.5), a shorter time window (such as 150 ms) is allocated to improve computational efficiency. The time window allocation can be implemented by a priority queue algorithm to generate a dynamic optimization adjustment strategy, which is output in JSON format, including time window allocation and optimization priority.

[0109] It can be understood that this strategy adaptively adjusts the wiring efficiency (such as optimizing the cable path length) and space utilization (such as achieving a site utilization rate of 85%), providing guidance for the iterative optimization of the layout control module.

[0110] In some embodiments, for a deep learning-based intelligent optimization design method for three-dimensional layout of a substation, a specific implementation is provided, focusing on inputting the dynamic optimization adjustment strategy into the layout control module to generate an optimized three-dimensional layout path through iterative optimization, giving priority to electrical safety distance and maintenance accessibility.

[0111] Specifically, the dynamic optimization adjustment strategy is converted into a control vector, and the timing information of the three-dimensional layout sequence and the substation operation constraints are superimposed.

[0112] It can be appreciated that the dynamic optimization adjustment strategy is generated by claim 7, containing optimization time window allocation and priority information, for adaptive adjustment of wiring efficiency and space utilization in substation layout.

[0113] The dynamic optimization adjustment strategy is input into the layout control module in JSON format, converted into a control vector, such as a 128-dimensional vector, representing optimization priorities and time window allocation. The timing information of the three-dimensional layout sequence (such as the update frequency of device coordinates) and the operation constraints of the substation (such as the minimization of cable wiring length, the maintenance channel width not less than 1.5 meters) can be superimposed through a data processing pipeline.

[0114] It can be appreciated that the control vector integrates time window allocation (such as 800ms optimization time for high-voltage device area) and operation constraints (such as electrical safety distance not less than 2.5 meters), providing comprehensive guidance for subsequent optimization.

[0115] Further, the control vector is processed by a multi-layer residual network and an attention mechanism to generate a preliminary layout path, which prioritizes electrical safety distance and maintenance channel accessibility.

[0116] Specifically, the control vector is input into a combined model of multi-layer residual network (ResNet) and attention mechanism, running in a deep learning framework (such as TensorFlow).

[0117] ResNet adopts an 18-layer residual structure to process the control vector, preserving complex constraint details such as device spacing and terrain adaptability. The attention mechanism (based on a self-attention model) assigns weights, prioritizing electrical safety distance (such as 2.5 meters between high-voltage circuit breakers) and maintenance channel accessibility (such as 1.5 meters channel width). For example, the attention mechanism enhances the weight of high-temperature device areas, adjusting their spacing with adjacent devices to meet heat dissipation requirements. The model can be trained by an Adam optimizer to generate a preliminary layout path, output as a vector sequence containing device coordinates and channel layout.

[0118] It can be appreciated that the preliminary layout path prioritizes meeting the electrical safety and maintenance accessibility requirements of the substation.

[0119] Further, the preliminary layout path is iteratively optimized by a feedback network to generate a final substation three-dimensional layout path.

[0120] Specifically, the preliminary layout path is input into the feedback network, which adopts a recurrent neural network (RNN) structure (such as 2 layers of GRU, 128 neurons per layer) for iterative optimization.

[0121] Further, the feedback network analyzes the performance of the preliminary layout path, such as checking whether the electrical safety distance meets the specification, and whether the maintenance channel is accessible. The device coordinates and the wiring path can be adjusted through multiple iterations (e.g., 15 times) to optimize the space utilization (e.g., to 85%) and the wiring efficiency (e.g., to reduce the cable length by 10%).

[0122] Understandably, the feedback network outputs the final substation three-dimensional layout path in a BIM (Building Information Modeling) format file, which contains the final coordinates of the devices, the wiring path, and the layout of the maintenance channel, and meets the operation and maintenance requirements of the substation.

[0123] The above description is only an exemplary embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structural transformation made by using the content of the present application specification and drawings, or direct / indirect application in other related technical fields is included in the patent protection scope of the present application.

Claims

1. A deep learning-based intelligent optimization design method for the three-dimensional layout of substations, characterized in that, This method, applied to the spatial layout optimization of substation equipment, adaptively processes complex substation constraints using a deep learning model to generate a three-dimensional layout scheme that balances space utilization, electrical safety distances, and maintenance accessibility. Specifically, it includes: Based on the received substation design requirements, the spatial characteristics of the equipment layout are collected through the 3D layout generation module. The spatial characteristics represent the geometric distribution, electrical safety distance and terrain constraints of the equipment in 3D space. The data coordination module coordinates the data interaction characteristics between the substation equipment data and the 3D layout generation module, wherein the data interaction characteristics characterize the dynamic mapping relationship between equipment electrical parameters, thermal distribution and layout constraints. The spatial features and the data interaction features are input into the deep learning optimization module. The deep learning optimization module fuses the spatial features and the data interaction features through an adaptive deep neural network to generate an optimized three-dimensional layout scheme that meets the space utilization and maintenance accessibility requirements of the substation.

2. The method as described in claim 1, characterized in that, The method further includes: The feature extraction module for spatial feature changes is processed by deep learning to generate a first layout feature set, wherein the output of the first layout feature set represents the spatial feature set of the three-dimensional layout generation module, including device spacing and terrain adaptability. The feature extraction module for data interaction feature changes is processed by deep learning to generate a second interaction feature set, wherein the output of the second interaction feature set represents the interaction feature set of the data coordination module, including the dynamic correlation between electrical parameters and thermal distribution. A three-dimensional layout optimization depth model for substations is constructed based on the first layout feature set and the second interaction feature set. The three-dimensional layout optimization depth model for substations optimizes the substation layout by adaptively fusing the spatial features and data interaction features.

3. The method as described in claim 2, characterized in that, The step of acquiring the spatial features of the device layout through the 3D layout generation module includes: Determine the layout change data generated by the first layout feature set according to design requirements, wherein the layout change data includes the geometric vector of equipment spatial distribution, electrical safety distance weight and terrain constraint factor; Based on the layout change data, the feature intensity of each 3D layout generation module is calculated using a convolutional neural network. The feature intensity reflects the spatial adaptability of the device layout. The spatial features are generated based on the feature intensity.

4. The method as described in claim 2, characterized in that, The step of coordinating the data interaction features between the substation equipment data and the 3D layout generation module through the data coordination module includes: Determine the data interaction change data generated by the second interaction feature set according to the design requirements, wherein the data interaction change data includes the transmission amplitude of the device electrical parameters, the heat distribution weight, and the data flow direction factor; The transmission amplitude, heat distribution weight, and flow direction factor of each data coordination module are converted into data interaction features using a recurrent neural network.

5. The method as described in claim 1, characterized in that, The deep learning optimization module fuses the spatial features and data interaction features through an adaptive deep neural network, including the following steps: The spatial features sent in real time by the 3D layout generation module and the data interaction features sent by the data coordination module are obtained. An optimized 3D layout scheme is generated by fusing the spatial features and the data interaction features through an adaptive deep neural network. The scheme optimizes the layout by dynamically adjusting the electrical safety distance and the width of the maintenance passage.

6. The method as described in claim 5, characterized in that, The step of fusing the spatial features and the data interaction features through an adaptive deep neural network includes: The spatial features and the data interaction features are input into the multilayer perceptron to generate a fused feature vector, which represents the comprehensive constraints of the substation layout. The fused feature vectors are processed by a residual neural network and an attention mechanism to generate an optimized 3D layout sequence that prioritizes electrical safety and maintenance accessibility.

7. The method as described in claim 5, characterized in that, The step of generating the optimized 3D layout scheme includes: Based on the complexity of the three-dimensional layout sequence, an optimization time window is allocated to generate a dynamic optimization adjustment strategy. The dynamic optimization adjustment strategy adaptively adjusts the wiring efficiency and space utilization in the substation layout. The dynamic optimization and adjustment strategy is input into the layout control module, and an optimized 3D layout path is generated through iterative optimization.

8. The method as described in claim 7, characterized in that, The step of allocating and optimizing time windows based on the complexity of the three-dimensional layout sequence includes: The three-dimensional layout sequence is subjected to feature extraction by a multi-scale convolutional neural network to obtain the first-scale feature S1 and the second-scale feature S2, which respectively characterize the local electrical constraints and global spatial complexity of the three-dimensional layout sequence. According to the formula Calculate the relative importance distribution, where R represents the relative importance distribution, S1 and S2 represent the complexity of the first and second scale features respectively, w1 and w2 are weight coefficients, and ∈ is a small constant to prevent division by zero; The optimal time window allocation method is determined based on the relative importance distribution, and the dynamic optimization adjustment strategy is generated.

9. The method as described in claim 8, characterized in that, The step of determining the optimal time window allocation method based on the relative importance distribution includes: The dynamic threshold T corresponding to the relative importance distribution is generated by an adaptive filter, where T = σ(R), and σ is the Sigmoid function used to normalize the relative importance distribution; The allocation factor F is calculated based on the relative importance distribution and the dynamic threshold, where F represents the allocation factor; The allocation method of the optimization time window is determined based on the allocation factor. When F is greater than the preset threshold, a longer time window is allocated to optimize the complex layout. Otherwise, a shorter time window is allocated to improve the computational efficiency, thereby generating the dynamic optimization adjustment strategy.

10. The method as described in claim 7, characterized in that, The step of inputting the dynamic optimization adjustment strategy into the layout control module and generating an optimized 3D layout path through iterative optimization includes: The dynamic optimization and adjustment strategy is converted into a control vector, and the timing information of the three-dimensional layout sequence and the substation operation constraints are superimposed on it. The control vector is processed by a multi-layer residual network and an attention mechanism to generate a preliminary layout path, which prioritizes electrical safety distance and maintenance accessibility. The initial layout path is iteratively optimized using a feedback network to generate the final three-dimensional layout path of the substation.

Images