Machine Learning-Based Optimization Method for Internal Wiring of Integrated Power Supply Boxes

Abstract

This invention discloses a machine learning-based method for optimizing internal wiring in an integrated power supply box, comprising the following steps: Step 1: Constructing a spatial topology map; Step 2: Generating a three-dimensional wiring path potential tensor using a diffusion transformer combined with an attention mechanism; Step 3: Forming a historical wiring path record set; Step 4: Identifying electromagnetic interference (EMI) sensitive areas and constructing a reverse electromagnetic backtracking disturbance field based on the EMI sensitive areas and the historical wiring path record set, detecting and extracting a subset of interference paths; Step 5: Performing local rerouting on the subset of interference paths to obtain a wiring path correction set and marking sensitive path segments; Step 6: Constructing a virtual impedance model based on the sensitive path segments to determine easily coupled points; Step 7: Inserting virtual impedance nodes at the easily coupled points of the sensitive path segments, ultimately forming a set of optimized wiring paths. This invention integrates diffusion modeling and reverse electromagnetic backtracking to achieve intelligent optimization of wiring in integrated power supply boxes.

Description

Technical Field

[0001] This invention relates to the fields of electromagnetic compatibility and wiring optimization technology, and in particular to a method for optimizing wiring inside an integrated power supply box based on machine learning. Background Technology

[0002] With the rapid development of new energy vehicles, rail transit, and intelligent power distribution systems towards higher power density and higher integration, integrated power supply boxes, as core energy scheduling and distribution units, have a crucial impact on the stable operation of the entire vehicle or equipment system due to their internal wiring space utilization, electromagnetic compatibility, and manufacturing feasibility. Existing power supply box wiring design methods mainly rely on manual experience combined with rule-driven wiring algorithms for initial layout and interference avoidance, but in practical engineering applications, the following problems are commonly encountered:

[0003] In highly integrated spatial structures, electronic devices and terminal interfaces are arranged in complex patterns with dense routing paths. Existing methods struggle to accurately model the topological relationships of devices, interfaces, and routing constraints in three-dimensional space, leading to uneven path distribution and severe overlap in the initial routing stage. Traditional routing methods based on heuristic search or graph theory cannot dynamically adjust based on historical routing data and the potential of complex routing, easily resulting in high-risk routing areas such as signal frequency overlap and current coupling overlap. For the identification of interference areas and the selection of optimized paths, most methods rely only on simplified electromagnetic compatibility rules or distance constraints, lacking modeling of the dynamic interference propagation mechanism between paths, making it impossible to accurately trace the source of interference and guide routing reconstruction. At the same time, existing optimization strategies lack insertable anti-interference node design mechanisms at high-frequency signal coupling points, making it difficult to effectively passivate and adjust the structure of local coupling hotspots. This results in significant fluctuations in the measured electromagnetic interference indicators of the final routing scheme, as well as insufficient manufacturability and robustness.

[0004] Therefore, how to provide a machine learning-based method for optimizing the internal wiring of an integrated power supply box is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a machine learning-based method for optimizing the internal wiring of an integrated power supply box. This invention integrates a diffusion converter and an attention mechanism to construct a three-dimensional wiring path potential tensor, and combines a reverse electromagnetic backtracking perturbation mechanism and a virtual impedance model to achieve intelligent generation, interference path identification, and local anti-interference enhancement of the internal wiring of the integrated power supply box. It has comprehensive advantages such as high wiring efficiency, strong electromagnetic compatibility, high precision in coupling interference control, and excellent manufacturability, and is suitable for the automated optimization design of high-density wiring structures in complex spaces.

[0006] The machine learning-based integrated power supply box internal wiring optimization method according to the present invention includes the following steps:

[0007] Step 1: Create a 3D spatial model of the electronic components, terminal interfaces, and connection requirements within the integrated power supply box, and construct a spatial topology diagram;

[0008] Step 2: Based on the aforementioned spatial topology graph, a three-dimensional wiring path potential tensor is generated using a diffusion transformer combined with an attention mechanism;

[0009] Step 3: Based on the three-dimensional wiring path potential tensor, a wiring path set is initially generated, and the routing trajectory, electrical parameters and wiring sequence of each wiring path are recorded to form a wiring path history record set;

[0010] Step 4: Identify electromagnetic interference sensitive areas, and construct a reverse electromagnetic backtracking disturbance field based on the electromagnetic interference sensitive areas and the wiring path historical record set, and detect and extract the interference path subset whose overall disturbance score exceeds the preset upper limit threshold;

[0011] Step 5: Perform local rerouting on the interference path subset to obtain the wiring path correction set after reverse electromagnetic backtracking optimization, and mark the sensitive path segments that still exist in the electromagnetic interference sensitive area.

[0012] Step 6: Construct a virtual impedance model based on the sensitive path segment, simulate the electromagnetic response characteristics of the sensitive path segment at different frequencies, and determine the easily coupled points;

[0013] Step 7: Insert virtual impedance nodes at easily coupled points in the sensitive path segment, and enhance the anti-interference capability of the path through local layout adjustments, ultimately forming a set of optimized cabling paths.

[0014] Optionally, step one: performing three-dimensional spatial modeling of the electronic components, terminal interfaces, and connection requirements within the integrated power supply box, and constructing a spatial topology diagram, specifically:

[0015] Each electronic component in the integrated power supply box is modeled in three dimensions according to its physical size, mounting plane, and mounting orientation, and a unique spatial coordinate system origin and orientation vector are assigned to each electronic component.

[0016] The position coordinates, wiring direction, connector type, pin arrangement and corresponding signal type of each terminal interface are structured and encoded, and a local spatial constraint relationship between the terminal interface and electronic device is established based on the terminal interface.

[0017] Analyze the electrical connection requirements between various electronic devices, and construct directed connection edges based on the starting node and the target node in the connection pair;

[0018] Based on the electronic device nodes, terminal interface nodes, and connection edges completed in 3D space modeling, construct a spatial topology diagram.

[0019] Optionally, the generation of the three-dimensional wiring path potential tensor using a diffusion transformer combined with an attention mechanism specifically involves:

[0020] The spatial topology graph is input into the diffusion transformer, and a discrete time-step diffusion mechanism is used to perform diffusion operations on the node features. The diffusion operation includes, at each diffusion time step, propagating the current node features along the edge connection direction to the neighboring nodes through the graph convolution kernel. An attention mechanism is introduced in each diffusion time step, and the corresponding attention score is calculated based on the spatial Euclidean distance between nodes, the signal frequency difference, the current level difference, and the overlap rate of the channel. The attention weight matrix is ​​formed by normalization and used to dynamically weight and update the diffusion direction of each neighboring node.

[0021] Repeat the diffusion operation until the preset maximum number of iterations is reached. The outputs of all diffusion time steps are accumulated and superimposed at the channel level to form a potential stacking tensor. The attention weight matrix is ​​then applied to the potential stacking tensor using a dot-multiplication weighting method to form a three-dimensional wiring path potential tensor.

[0022] Optionally, step three: Based on the three-dimensional wiring path potential tensor, a wiring path set is initially generated, and the routing trajectory, electrical parameters, and wiring sequence of each wiring path are recorded to form a wiring path history record set, specifically:

[0023] For each potential channel corresponding to a wiring connection pair in the three-dimensional wiring path potential tensor, a greedy search strategy is used to extract the path starting from the starting node of the connection pair.

[0024] The greedy search strategy includes calculating the potential value of adjacent grid points in the three-dimensional spatial grid, starting from the location of the starting node, and selecting the grid point with the largest potential value as the next routing position in the current step. The potential value is the probability score of the spatial grid point corresponding to the starting node to the target node of the current routing connection pair in the three-dimensional routing path potential tensor. The probability score is determined by the diffusion transformer propagating the node features through multiple discrete-time diffusion steps and dynamically weighting and updating it through the attention mechanism, which represents the relative strength of the feasibility of the routing channel at the current spatial location point.

[0025] Mark the currently selected grid point as a visited node, and proceed step by step until the target node of the corresponding connection pair is reached, thus completing the search for an initial wiring path;

[0026] During the path search process, the spatial coordinate sequence of each hop is recorded to form a complete path trajectory;

[0027] Based on the information of the electronic devices connected to each wiring path, the current level, voltage level and signal frequency carried by the wiring path are extracted and associated to obtain the electrical parameters corresponding to the wiring path;

[0028] The routing path, electrical parameters, and the generated wiring sequence during the search process are numbered and archived to form a wiring path history record set.

[0029] Optionally, the identification of electromagnetic interference sensitive areas specifically includes:

[0030] Map the routing trajectory of each routing path in the routing path history set to a three-dimensional spatial mesh to construct a spatial routing distribution map;

[0031] In the three-dimensional spatial grid, the wiring path superposition density, signal frequency overlap, and total current intensity superposition of each spatial grid cell are statistically analyzed, and the minimum spacing between adjacent wiring paths is measured to construct an electromagnetic interference characteristic matrix.

[0032] Spatial grid cells that meet any of the conditions are marked as electromagnetic interference sensitive areas, and corresponding spatial identification indexes are established:

[0033] Condition 1: The overlap density of wiring paths exceeds a preset density threshold;

[0034] Condition 2: More than 3 wiring paths exhibit signal frequency overlap within a spatial range;

[0035] Condition 3: The minimum spacing between adjacent wiring paths is lower than the manufacturing tolerance threshold;

[0036] Condition 4: The total intensity of the path current exceeds the preset environmental current limiting threshold.

[0037] Optionally, the step of constructing a reverse electromagnetic backtracking disturbance field based on the electromagnetic interference sensitive area and the historical data set of the wiring path, and detecting and extracting a subset of interference paths whose overall disturbance score exceeds a preset upper limit threshold, specifically involves:

[0038] The three-dimensional routing trajectories and corresponding electrical parameters of each wiring path in the wiring path history record set are mapped to the spatial grid where the electromagnetic interference sensitive area is located, and the path segments that cross any electromagnetic interference sensitive area in all wiring paths are selected as the starting point of disturbance propagation.

[0039] Define a three-dimensional perturbation score tensor consistent with the wiring space structure to represent the perturbation intensity distribution of each spatial grid cell, and set the initial value to zero;

[0040] For each path segment that traverses an area sensitive to electromagnetic interference, the grid position where the path segment is located is taken as the starting node, and the path segment is traversed back segment by segment towards the starting node in reverse order of the coordinate index of the path segment in the wiring path history record set.

[0041] At each path segment, the disturbance contribution value is determined by combining the current intensity of the path segment, the signal frequency, and the hop distance from the electromagnetic interference sensitive area. Specifically, the current intensity of the path segment is multiplied by the signal frequency, and then multiplied by a weighting coefficient that decreases exponentially with the number of hops, so that the larger the number of hops, the smaller the disturbance contribution value.

[0042] Write the disturbance contribution value of each path segment into the corresponding three-dimensional spatial grid position, and accumulate the existing disturbance contribution value at the three-dimensional spatial grid position until all wiring paths have completed back propagation to obtain the reverse electromagnetic back propagation disturbance field.

[0043] Based on the routing trajectory of each routing path in the historical routing history set, the disturbance contribution value of the routing path at the corresponding three-dimensional spatial grid position in the reverse electromagnetic backtracking disturbance field is extracted and integrated to form the overall disturbance score of the routing path.

[0044] Wiring paths whose overall disturbance score exceeds a preset upper limit threshold are selected as a subset of interference paths.

[0045] Optionally, performing local rerouting on the subset of interfering paths specifically involves:

[0046] The three-dimensional trace of each path in the interference path subset is analyzed segment by segment to determine whether the local disturbance score in the reverse electromagnetic backtracking disturbance field exceeds the preset local interference threshold.

[0047] If the local disturbance score of the path segment is lower than or equal to the preset local disturbance threshold, the original path remains unchanged.

[0048] If the local disturbance score of a path segment is higher than the preset local interference threshold, a local rerouting is performed. The local rerouting is to re-perform a local greedy search in the three-dimensional cabling path potential tensor to find a new path segment with the highest potential value that does not cross an electromagnetic interference sensitive area. If found, the corresponding path segment in the original path is replaced with the new path segment to obtain the cabling path correction set. Otherwise, the path segment is marked as a sensitive path segment.

[0049] Optionally, step six: constructing a virtual impedance model based on the sensitive path segment, simulating the electromagnetic response characteristics of the sensitive path segment at different frequencies, and determining easily coupled points, specifically involves:

[0050] The three-dimensional routing structure and corresponding electrical parameters of the sensitive path segment are input into the electromagnetic field simulation environment. Distributed parameters composed of series and parallel combinations of resistors, inductors and capacitors are established between the two end nodes of the sensitive path segment to form a virtual impedance model of the sensitive path segment.

[0051] The virtual impedance model is used to perform impedance-frequency sweep simulation on the sensitive path segment over a wide frequency range to obtain the impedance-frequency characteristic curves of the sensitive path segment at different frequency points.

[0052] Based on the impedance-frequency characteristic curve, the location where the impedance changes more than a preset change threshold within a unit frequency range near the target operating frequency point and is prone to electromagnetic coupling with adjacent paths is identified as the easily coupled point of the sensitive path segment.

[0053] Optionally, step seven involves inserting virtual impedance nodes at easily coupled points in the sensitive path segment. This enhances the path's anti-interference capability through local layout adjustments, ultimately forming a set of optimized cabling paths. Specifically:

[0054] Virtual impedance nodes are inserted at the easily coupled points of each sensitive path segment. The virtual impedance nodes are equivalent impedance units, which are composed of inductors, capacitors and resistors connected in series. The equivalent impedance value is based on the minimum impedance value in the frequency range where impedance changes occur near the target operating frequency point in the virtual impedance model. It is used to introduce anti-phase impedance matching at the easily coupled points to attenuate the coupling interference energy propagating along the frequency point.

[0055] Based on the local remaining empty space in the wiring space and the structural assembly constraints, the position of the virtual impedance node is adjusted by displacement, and the insertion point is selected in the metal shell shielding edge, dielectric insulation gap or low frequency interference zone to avoid adding new coupling hot spots.

[0056] After the virtual impedance node is injected, the overall routing structure of the sensitive path segment is partially reconstructed and adjusted, including routing angle optimization, path segment stretching / compression, and redistribution of minimum spacing with adjacent lines, so as to improve the anti-interference space separation between the wiring without violating the original electrical connection.

[0057] All sensitive path segments that have undergone anti-interference enhancement processing are merged with the routing path correction set to output the final routing optimized path set.

[0058] The beneficial effects of this invention are:

[0059] This invention addresses the problems of low 3D spatial modeling accuracy, irreversible path interference tracing, and difficulty in actively suppressing local coupling in integrated power supply box wiring by synergistically integrating a diffusion converter and an attention mechanism. It proposes a full-process wiring optimization method driven by the wiring path potential tensor. In the wiring generation stage, a graph-structured diffusion propagation mechanism is used to dynamically model the path potential in the spatial topology, and an attention module based on spatial distance, current level, and frequency difference weighting is introduced to achieve probabilistic modeling of the routing trends of different wiring connection pairs in the 3D mesh. During path extraction, a greedy search strategy is combined to perform channel-level decoding of the potential tensor, constructing an efficient initial wiring path set that conforms to manufacturing rules. For electromagnetic interference areas that are difficult to identify in advance in high-density spaces, this invention introduces historical wiring data and spatial density parameters to construct an electromagnetic interference feature matrix, and simulates the spatial intensity distribution of interference sources propagating backward along the path based on a reverse electromagnetic backtracking disturbance field mechanism, identifying interference paths whose overall disturbance score exceeds a set threshold. Furthermore, by establishing a multi-frequency domain virtual impedance model for sensitive path segments, the spectrum identification of electromagnetic coupling response and extraction of easily coupled points are achieved. Finally, bandwidth-controlled virtual impedance nodes are injected into local areas with high coupling risk. Through spatial structure fine-tuning and impedance matching, source domain attenuation and structural passivation of coupling interference are achieved. This application can significantly improve the anti-interference performance and manufacturability of cabling schemes in complex spaces, providing a full-process, data-driven technical path for integrated power supply box cabling automation and electromagnetic compatibility optimization. Attached Figure Description

[0060] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0061] Figure 1 This is an overall flowchart of the machine learning-based integrated power supply box internal wiring optimization method proposed in this invention.

[0062] Figure 2 This is a flowchart illustrating the path extraction process of the machine learning-based integrated power supply box internal wiring optimization method proposed in this invention, which uses a greedy search based on the wiring path potential tensor to generate wiring paths.

[0063] Figure 3 This diagram illustrates the process of constructing the reverse electromagnetic backtracking perturbation field and generating path perturbation scores for the machine learning-based integrated power supply box internal wiring optimization method proposed in this invention. Detailed Implementation

[0064] Example 1:

[0065] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0066] refer to Figures 1-3 The machine learning-based method for optimizing the internal wiring of an integrated power supply box includes the following steps:

[0067] Step 1: Create a 3D spatial model of the electronic components, terminal interfaces, and connection requirements within the integrated power supply box, and construct a spatial topology diagram;

[0068] Step 2: Based on the aforementioned spatial topology graph, a three-dimensional wiring path potential tensor is generated using a diffusion transformer combined with an attention mechanism;

[0069] Step 3: Based on the three-dimensional wiring path potential tensor, a wiring path set is initially generated, and the routing trajectory, electrical parameters and wiring sequence of each wiring path are recorded to form a wiring path history record set;

[0070] Step 4: Identify electromagnetic interference sensitive areas, and construct a reverse electromagnetic backtracking disturbance field based on the electromagnetic interference sensitive areas and the wiring path historical record set, and detect and extract the interference path subset whose overall disturbance score exceeds the preset upper limit threshold;

[0071] Step 5: Perform local rerouting on the interference path subset to obtain the wiring path correction set after reverse electromagnetic backtracking optimization, and mark the sensitive path segments that still exist in the electromagnetic interference sensitive area.

[0072] Step 6: Construct a virtual impedance model based on the sensitive path segment, simulate the electromagnetic response characteristics of the sensitive path segment at different frequencies, and determine the easily coupled points;

[0073] Step 7: Insert virtual impedance nodes at easily coupled points in the sensitive path segment, and enhance the anti-interference capability of the path through local layout adjustments, ultimately forming a set of optimized cabling paths.

[0074] This invention constructs a three-dimensional spatial topology map of the integrated power supply box, combines a diffusion converter and an attention mechanism to establish a three-dimensional wiring path potential tensor, and on this basis, completes the entire process optimization operation of initial wiring, interference path identification, path disturbance correction, and anti-interference enhancement, realizing a deep integration of intelligent wiring generation and electromagnetic interference control. Compared with traditional rule-based wiring algorithms, this invention has the advantages of high spatial modeling accuracy, strong wiring feasibility prediction capability, strong interference control path backtracking capability, and refined and pluggable anti-interference processing. It effectively solves the problems of multi-path superposition, current frequency intersection, and high spatial coupling sensitivity in complex environments, significantly improving the electromagnetic compatibility, manufacturability, and overall operational reliability of power supply box wiring.

[0075] In this embodiment, step one, which involves creating a three-dimensional spatial model of the electronic components, terminal interfaces, and connection requirements within the integrated power supply box, and constructing a spatial topology diagram, specifically includes:

[0076] Each electronic component in the integrated power supply box is modeled in three dimensions according to its physical size, mounting plane, and mounting orientation, and a unique spatial coordinate system origin and orientation vector are assigned to each electronic component.

[0077] The position coordinates, wiring direction, connector type, pin arrangement and corresponding signal type of each terminal interface are structured and encoded, and a local spatial constraint relationship between the terminal interface and electronic device is established based on the terminal interface.

[0078] Analyze the electrical connection requirements between various electronic devices, and construct directed connection edges based on the starting node and the target node in the connection pair;

[0079] Based on the electronic device nodes, terminal interface nodes, and connection edges completed in 3D space modeling, construct a spatial topology diagram.

[0080] By creating a 3D model of the structural information of electronic components and terminal interfaces, and establishing a spatial coordinate system and local constraints, the model accurately reflects the mounting direction, spatial position, and connection constraints between components, greatly improving the accuracy of the initial modeling. Simultaneously, connection requirements are parsed and constructed as directed edges, forming a structured spatial topology graph, providing a precise and quantifiable data foundation for wiring path modeling and generation. Compared to traditional manual wiring or simplified structural representations, this modeling method better suits the complex wiring requirements of high-density electronic systems, significantly improving the accuracy of path feasibility calculations and the scalability of algorithm training during the wiring phase.

[0081] In this embodiment, the generation of the three-dimensional wiring path potential tensor using a diffusion transformer combined with an attention mechanism specifically involves:

[0082] The spatial topology graph is input into the diffusion transformer, and a discrete time-step diffusion mechanism is used to perform diffusion operations on the node features. The diffusion operation includes, at each diffusion time step, propagating the current node features along the edge connection direction to the neighboring nodes through the graph convolution kernel. An attention mechanism is introduced in each diffusion time step, and the corresponding attention score is calculated based on the spatial Euclidean distance between nodes, the signal frequency difference, the current level difference, and the overlap rate of the channel. The attention weight matrix is ​​formed by normalization and used to dynamically weight and update the diffusion direction of each neighboring node.

[0083] ;

[0084] in, Represents a node To adjacent nodes Attention score Represents a node With nodes The Euclidean distance between them Represents a node With nodes The signal frequency difference between them Represents a node With nodes The difference in current levels between them Represents a node and nodes The overlap rate of the path channel, with a value range of [value missing]. 1 indicates complete overlap. and Indicates the weighting coefficient. Represents an exponential function. This represents the normalization factor, which makes the sum of the attention scores of all adjacent nodes equal to 1. , This indicates the absolute value operation. and The range of values ​​is ,in This represents the total number of nodes in the spatial topology graph;

[0085] Repeat the diffusion operation until the preset maximum number of iterations is reached. The outputs of all diffusion time steps are accumulated and superimposed at the channel level to form a potential stacking tensor. The attention weight matrix is ​​then applied to the potential stacking tensor using a dot-multiplication weighting method to form a three-dimensional wiring path potential tensor.

[0086] By combining a diffusion transformer with an attention mechanism to generate a 3D routing path potential tensor, a probability distribution of routing feasibility under different spatial locations is effectively established. The diffusion mechanism realizes the temporal propagation modeling of routing trends, while the attention mechanism dynamically perceives the combined effects of spatial distance, current level, and frequency difference to generate a high-resolution potential tensor. Compared with traditional static scoring models, this approach can dynamically adjust routing preferences, enhance the model's ability to identify high-quality paths in complex topologies, improve the directionality and convergence efficiency of routing search, and provide a stable and efficient probabilistic driving foundation for path generation.

[0087] In this embodiment, step three: Based on the three-dimensional wiring path potential tensor, a wiring path set is initially generated, and the routing trajectory, electrical parameters, and wiring sequence of each wiring path are recorded to form a wiring path history record set. Specifically:

[0088] For each potential channel corresponding to a wiring connection pair in the three-dimensional wiring path potential tensor, a greedy search strategy is used to extract the path starting from the starting node of the connection pair.

[0089] The greedy search strategy includes calculating the potential value of adjacent grid points in the three-dimensional spatial grid, starting from the location of the starting node, and selecting the grid point with the largest potential value as the next routing position in the current step. The potential value is the probability score of the spatial grid point corresponding to the starting node to the target node of the current routing connection pair in the three-dimensional routing path potential tensor. The probability score is determined by the diffusion transformer propagating the node features through multiple discrete-time diffusion steps and dynamically weighting and updating it through the attention mechanism, which represents the relative strength of the feasibility of the routing channel at the current spatial location point.

[0090] The potential value of this invention refers to the probability score result corresponding to a specific spatial location (a spatial grid point in the path search process) in the potential stacking tensor, that is, the value result of the potential stacking tensor in a specific dimension, which is a scalar.

[0091] Mark the currently selected grid point as a visited node, and proceed step by step until the target node of the corresponding connection pair is reached, thus completing the search for an initial wiring path;

[0092] During the path search process, the spatial coordinate sequence of each hop is recorded to form a complete path trajectory;

[0093] Based on the information of the electronic devices connected to each wiring path, the current level, voltage level and signal frequency carried by the wiring path are extracted and associated to obtain the electrical parameters corresponding to the wiring path;

[0094] The routing path, electrical parameters, and the generated wiring sequence during the search process are numbered and archived to form a wiring path history record set.

[0095] By combining the 3D routing path potential tensor with a greedy search strategy for path extraction, a set of routing paths that satisfy the current spatial constraints can be generated quickly and stably. The greedy search strategy selects the optimal routing point at each step based on the potential score, which can reduce the computational cost of the algorithm while maintaining path feasibility and a high-score trend, making it suitable for large-scale routing scenarios. Synchronous recording of routing trajectories and electrical parameters provides sufficient data support for subsequent interference identification and path optimization, significantly improving the automation level of the routing generation stage and the quality of the initial paths.

[0096] In this embodiment, the identification of electromagnetic interference sensitive areas specifically refers to:

[0097] Map the routing trajectory of each routing path in the routing path history set to a three-dimensional spatial mesh to construct a spatial routing distribution map;

[0098] In the three-dimensional spatial grid, the wiring path superposition density, signal frequency overlap, and total current intensity superposition of each spatial grid cell are statistically analyzed, and the minimum spacing between adjacent wiring paths is measured to construct an electromagnetic interference characteristic matrix.

[0099] Spatial grid cells that meet any of the conditions are marked as electromagnetic interference sensitive areas, and corresponding spatial identification indexes are established:

[0100] Condition 1: The overlap density of wiring paths exceeds a preset density threshold;

[0101] Condition 2: More than 3 wiring paths exhibit signal frequency overlap within a spatial range;

[0102] Condition 3: The minimum spacing between adjacent wiring paths is lower than the manufacturing tolerance threshold;

[0103] Condition 4: The total intensity of the path current exceeds the preset environmental current limiting threshold.

[0104] This application constructs a spatial electromagnetic interference feature matrix by mapping the wiring path trajectory and statistically analyzing multi-dimensional electromagnetic features (wiring path superposition density, signal frequency overlap, total superposition current intensity, and path spacing). This matrix can quickly identify potential interference areas after wiring is completed. It not only considers traditional spacing and frequency indicators but also combines superposition current and spatial overlap, improving the comprehensiveness and accuracy of electromagnetically sensitive area identification and providing clear and quantifiable input conditions for tracing and reconstructing interference paths.

[0105] In this embodiment, the step of constructing a reverse electromagnetic backtracking disturbance field based on the electromagnetic interference sensitive area and the historical data set of wiring paths, and detecting and extracting a subset of interference paths whose overall disturbance score exceeds a preset upper limit threshold, specifically involves:

[0106] The three-dimensional routing trajectories and corresponding electrical parameters of each wiring path in the wiring path history record set are mapped to the spatial grid where the electromagnetic interference sensitive area is located, and the path segments that cross any electromagnetic interference sensitive area in all wiring paths are selected as the starting point of disturbance propagation.

[0107] Define a three-dimensional perturbation score tensor consistent with the wiring space structure to represent the perturbation intensity distribution of each spatial grid cell, and set the initial value to zero;

[0108] For each path segment that traverses an area sensitive to electromagnetic interference, the grid position where the path segment is located is taken as the starting node, and the path segment is traversed back segment by segment towards the starting node in reverse order of the coordinate index of the path segment in the wiring path history record set.

[0109] At each path segment, the disturbance contribution value is determined by combining the current intensity of the path segment, the signal frequency, and the hop distance from the electromagnetic interference sensitive area. Specifically, the current intensity of the path segment is multiplied by the signal frequency, and then multiplied by a weighting coefficient that decreases exponentially with the number of hops, so that the larger the number of hops, the smaller the disturbance contribution value.

[0110] ;

[0111] in, Represents path segment The disturbance contribution value, Represents path segment The current intensity, Represents path segment The signal frequency, Represents path segment Minimum hop distance between the area and the electromagnetic interference sensitive area Indicates the exponential decay coefficient;

[0112] Write the disturbance contribution value of each path segment into the corresponding three-dimensional spatial grid position, and accumulate the existing disturbance contribution value at the three-dimensional spatial grid position until all wiring paths have completed back propagation to obtain the reverse electromagnetic back propagation disturbance field.

[0113] Based on the routing trajectory of each routing path in the historical routing history set, the disturbance contribution value of the routing path at the corresponding three-dimensional spatial grid position in the reverse electromagnetic backtracking disturbance field is extracted and integrated to form the overall disturbance score of the routing path.

[0114] Wiring paths whose overall disturbance score exceeds a preset upper limit threshold are selected as a subset of interference paths.

[0115] Constructing a reverse electromagnetic backtracking disturbance field enables full-process backtracking analysis of the interference source path. By using a disturbance integral scoring method, the contribution of each cabling path to the interference propagation link is quantitatively reflected. This invention introduces a physical modeling mechanism for disturbance field backtracking for the first time, realizing reverse reasoning from the interference result to the interference source path, significantly improving the source tracing capability and interpretability of interference identification. Compared with traditional static evaluation methods, this invention possesses stronger dynamic interference source detection and path correlation analysis capabilities, representing a significant innovation in the field of cabling interference modeling.

[0116] In this embodiment, performing local rerouting on the subset of interference paths specifically means:

[0117] The three-dimensional trace of each path in the interference path subset is analyzed segment by segment to determine whether the local disturbance score in the reverse electromagnetic backtracking disturbance field exceeds the preset local interference threshold.

[0118] If the local disturbance score of the path segment is lower than or equal to the preset local disturbance threshold, the original path remains unchanged.

[0119] If the local disturbance score of a path segment is higher than the preset local interference threshold, a local rerouting is performed. The local rerouting is to re-perform a local greedy search in the three-dimensional cabling path potential tensor to find a new path segment with the highest potential value that does not cross an electromagnetic interference sensitive area. If found, the corresponding path segment in the original path is replaced with the new path segment to obtain the cabling path correction set. Otherwise, the path segment is marked as a sensitive path segment.

[0120] By performing disturbance scoring analysis on interfering paths and executing path-level local rerouting operations based on the scoring distribution results, local interference hotspots can be accurately eliminated while ensuring the overall stability of the cabling structure. The rerouting process is based on the cabling path potential tensor, ensuring the high feasibility and anti-interference capability of the new path segments. This effectively reduces the large-scale structural damage caused by path reconstruction and achieves targeted optimization of locally high-interference segments. It is a highly robust and low-intrusion local cabling correction strategy.

[0121] In this embodiment, step six: constructing a virtual impedance model based on the sensitive path segment, simulating the electromagnetic response characteristics of the sensitive path segment at different frequencies, and determining easily coupled points, specifically includes:

[0122] The three-dimensional routing structure and corresponding electrical parameters of the sensitive path segment are input into the electromagnetic field simulation environment. Distributed parameters composed of series and parallel combinations of resistors, inductors and capacitors are established between the two end nodes of the sensitive path segment to form a virtual impedance model of the sensitive path segment.

[0123] The virtual impedance model is used to perform impedance-frequency sweep simulation on the sensitive path segment over a wide frequency range to obtain the impedance-frequency characteristic curves of the sensitive path segment at different frequency points.

[0124] Based on the impedance-frequency characteristic curve, the location where the impedance changes more than a preset change threshold within a unit frequency range near the target operating frequency point and is prone to electromagnetic coupling with adjacent paths is identified as the easily coupled point of the sensitive path segment.

[0125] The virtual impedance model is used to model and simulate the electromagnetic behavior of sensitive path segments at different frequencies without changing the actual cabling structure. The virtual impedance model constructs a distributed equivalent circuit by introducing equivalent resistors, inductors, and capacitors at both ends of the path, reflecting the impedance response characteristics of the sensitive path over a wide frequency range. The model parameters are set according to the actual electrical parameters of the path. Through frequency sweep simulation, it can accurately capture impedance anomalies, enhanced coupling, or resonance effects that may occur at specific frequency points. In the anti-interference enhancement step, the virtual impedance model provides key identification criteria for easily coupled points and is used to calculate the optimal matching parameters for inserting virtual impedance nodes, achieving directional suppression of coupling energy and minimization of inter-path interference, significantly improving the electromagnetic compatibility of the cabling system.

[0126] By constructing a virtual impedance model and performing impedance-frequency response simulations within the frequency spectrum, points on sensitive path segments with electromagnetic coupling risks can be accurately identified. The virtual impedance model maps wiring segments to a physically equivalent system composed of RLC elements, enabling response analysis in the frequency domain and providing a theoretical basis for impedance adjustment and structural optimization. Compared to empirical judgment or single-frequency analysis methods, it offers advantages such as wider frequency spectrum coverage and higher simulation accuracy, improving the accuracy of identifying easily coupled points and the scientific basis of anti-interference layout.

[0127] In this embodiment, step seven: inserting virtual impedance nodes at easily coupled points in the sensitive path segment, and enhancing the anti-interference capability of the path through local layout adjustments, ultimately forming a set of optimized wiring paths, specifically:

[0128] Virtual impedance nodes are inserted at the easily coupled points of each sensitive path segment. The virtual impedance nodes are equivalent impedance units, which are composed of inductors, capacitors and resistors connected in series. The equivalent impedance value is based on the minimum impedance value in the frequency range where impedance changes occur near the target operating frequency point in the virtual impedance model. It is used to introduce anti-phase impedance matching at the easily coupled points to attenuate the coupling interference energy propagating along the frequency point.

[0129] Based on the local remaining empty space in the wiring space and the structural assembly constraints, the position of the virtual impedance node is adjusted by displacement, and the insertion point is selected in the metal shell shielding edge, dielectric insulation gap or low frequency interference zone to avoid adding new coupling hot spots.

[0130] After the virtual impedance node is injected, the overall routing structure of the sensitive path segment is partially reconstructed and adjusted, including routing angle optimization, path segment stretching / compression, and redistribution of minimum spacing with adjacent lines, so as to improve the anti-interference space separation between the wiring without violating the original electrical connection.

[0131] All sensitive path segments that have undergone anti-interference enhancement processing are merged with the routing path correction set to output the final routing optimized path set.

[0132] Virtual impedance nodes are introduced at identified points of easy coupling. By inserting these nodes into the physical layer, the path impedance can be adjusted to create anti-phase interference or bandwidth suppression, effectively reducing electromagnetic interference propagation between coupled paths. The designed equivalent impedance nodes are parameter-adjustable, and their insertion positions are optimized based on spatial constraints to avoid creating new coupling hotspots. Finally, the enhanced and corrected paths are integrated, significantly improving the overall electromagnetic compatibility performance and stability of the cabling structure.

[0133] Example 2:

[0134] To verify the feasibility of this invention in practice, it was applied to the task of optimizing the wiring design of a high-density integrated power supply box in a certain intelligent transportation control system.

[0135] The power supply box has a compact structure, integrating dozens of electronic modules and hundreds of connection pairs. The wiring paths are dense and interwoven, with high current signal strength and a wide frequency range, making it prone to electromagnetic coupling interference in a confined space. During system wiring, problems such as abnormal control signals, false triggering of equipment, or communication link interference often occur due to unreasonable distribution of interference paths. Long-term reliance on manual adjustments is not only inefficient but also fails to meet the requirements of high reliability scenarios.

[0136] In this embodiment, the electronic components and terminal interfaces inside the power supply box are first modeled in three dimensions, and the connection topology between the components is automatically established. Based on this, a diffusion transformer combined with an attention mechanism is used to construct a three-dimensional wiring path potential tensor, and the routing feasibility of each region in the wiring space is modeled and evaluated. According to the potential tensor, the system automatically generates a set of shortest paths for each pair of connections using a greedy search strategy, recording their spatial trajectory and related electrical parameters to form a wiring path history record.

[0137] Next, the system performs joint analysis on multiple indicators, including densely overlapping areas of wiring paths, signal frequency resonance areas, and areas where the path spacing is less than a threshold, to construct an electromagnetic interference feature matrix and automatically identify multiple electromagnetic interference sensitive areas. Based on this, a reverse electromagnetic backtracking disturbance field is established by combining historical path information to trace the source segment contribution of each interference path, and scores and sorts it accordingly, automatically selecting a subset of interference paths.

[0138] The system performs local path disturbance correction on the above high-risk path subset, and repairs most of the interfering path segments by combining path fine-tuning and rerouting. For some residual path segments that still cross high-risk areas, it further identifies frequency response mutation points based on virtual impedance modeling and inserts equivalent impedance nodes to significantly reduce their electromagnetic impact on surrounding paths.

[0139] Table 1 Comparison of wiring interference indicators before and after power supply box optimization

[0140]

[0141] As can be seen from Table 1 above, the wiring optimization method proposed in this invention is significantly effective in reducing electromagnetic interference. Firstly, the average electromagnetic interference score per path decreased from 0.76 before optimization to 0.23, a reduction of 69.7%, fully demonstrating the excellent performance of the introduced diffusion potential modeling and interference backtracking mechanism in identifying and avoiding global interference distribution. The communication bit error rate also decreased from... Down to The reduction rate reached 98.7%, indicating a significant reduction in coupling interference between signal paths and effectively ensuring the stability of critical signal transmission. At the physical level, the number of overlapping paths within the electromagnetic interference-sensitive area decreased from 4.8 to 1.6, and the minimum path spacing increased from 2.2 mm to 4.5 mm, significantly improving the spatial distribution rationality and manufacturing feasibility. Furthermore, by inserting virtual impedance nodes at easily coupled points, the peak electromagnetic amplitude of the coupling was reduced by 61.2%, further verifying the structural advantages of this invention in interference suppression.

[0142] During the experiment, electromagnetic field simulation tools were used to perform electromagnetic coupling analysis on the wiring paths of the integrated power supply box before and after wiring optimization. The location with the strongest coupling interference among all sensitive path segments was selected as the evaluation point, and its peak electromagnetic field amplitude at the target operating frequency was recorded. Let the peak electromagnetic amplitude before optimization be... The optimized amplitude is The formula for calculating the reduction rate is:

[0143] ;

[0144] Taking the experimentally measured data as an example, the average peak value obtained in the test sample before optimization was 5.21V / m, and after optimization it was 2.02V / m. Substituting these values ​​into the above formula, the calculation is as follows:

[0145] .

[0146] Table 2 Comparison of routing path efficiency and convergence before and after optimization

[0147] project Before optimization After optimization Comparison results Initial wiring time (seconds) 327.4s 89.5s ↓72.7% Local disturbance adjustment of average iteration rounds 5 wheels 2 rounds ↓60% Cabling path ratio that meets design standards 78.3% 96.4% ↑23.1% Number of space-manufacturing conflicts 17 2 ↓88.2%

[0148] As can be further seen from Table 2 above, this invention also has significant advantages in terms of wiring efficiency and path search stability. The initial wiring time is reduced from 327.4 seconds to 89.5 seconds, a reduction of 72.7%, significantly improving the wiring generation speed, especially suitable for the rapid wiring requirements of high-density integration scenarios. The average number of iterations for local interference adjustment is reduced from 5 to 2, indicating that the wiring path already has better electromagnetic interference avoidance capabilities during initial generation, reducing the cost of subsequent adjustments. The proportion of wiring paths that meet design specifications (such as minimum spacing and electromagnetic compatibility) increases from 78.3% to 96.4%, improving the overall system compliance by 23.1%, demonstrating the system coordination and optimization capabilities from the algorithm to the structural level. Finally, the number of manufacturable conflicts in space is also significantly reduced from 17 to 2, a decrease of 88.2%, confirming that this method not only optimizes signal integrity but also considers the feasibility at the manufacturing level. This invention has practical value in improving electromagnetic compatibility, wiring intelligence, and engineering feasibility.

[0149] This embodiment fully verifies the effectiveness and advancement of the proposed machine learning-based integrated power supply box internal wiring optimization method in practical applications. By integrating 3D spatial modeling, diffusion converter potential guidance, reverse electromagnetic backtracking disturbance field construction, and virtual impedance enhancement mechanism, this application not only significantly reduces electromagnetic interference levels and improves signal transmission reliability, but also achieves comprehensive improvements in wiring efficiency, space utilization, and manufacturing feasibility. The optimized wiring path outperforms traditional methods in multiple key indicators such as interference score, bit error rate, and wiring time, fully demonstrating the invention's intelligent wiring capabilities in complex electromagnetic environments and its broad engineering application potential.

[0150] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A machine learning-based method for optimizing internal wiring in an integrated power supply box, characterized in that, Includes the following steps: Step 1: Create a 3D spatial model of the electronic components, terminal interfaces, and connection requirements within the integrated power supply box, and construct a spatial topology diagram; Step 2: Based on the aforementioned spatial topology graph, a three-dimensional wiring path potential tensor is generated using a diffusion transformer combined with an attention mechanism; Step 3: Based on the three-dimensional wiring path potential tensor, a wiring path set is initially generated, and the routing trajectory, electrical parameters and wiring sequence of each wiring path are recorded to form a wiring path history record set; Step 4: Identify electromagnetic interference sensitive areas, and construct a reverse electromagnetic backtracking disturbance field based on the electromagnetic interference sensitive areas and the wiring path historical record set, and detect and extract the interference path subset whose overall disturbance score exceeds the preset upper limit threshold; Step 5: Perform local rerouting on the interference path subset to obtain the wiring path correction set after reverse electromagnetic backtracking optimization, and mark the sensitive path segments that still exist in the electromagnetic interference sensitive area. Step 6: Construct a virtual impedance model based on the sensitive path segment, simulate the electromagnetic response characteristics of the sensitive path segment at different frequencies, and determine the easily coupled points; Step 7: Insert virtual impedance nodes at easily coupled points in the sensitive path segment, and enhance the anti-interference capability of the path through local layout adjustments, ultimately forming a set of optimized cabling paths; The method of generating the three-dimensional wiring path potential tensor using a diffusion transformer combined with an attention mechanism is as follows: The spatial topology graph is input into the diffusion transformer, and a discrete time-step diffusion mechanism is used to perform diffusion operations on the node features. The diffusion operation includes, at each diffusion time step, propagating the current node features along the edge connection direction to the neighboring nodes through the graph convolution kernel. An attention mechanism is introduced in each diffusion time step, and the corresponding attention score is calculated based on the spatial Euclidean distance between nodes, the signal frequency difference, the current level difference, and the overlap rate of the channel. The attention weight matrix is ​​formed by normalization and used to dynamically weight and update the diffusion direction of each neighboring node. Repeat the diffusion operation until the preset maximum number of iterations is reached. The outputs of all diffusion time steps are accumulated and superimposed at the channel level to form a potential stacking tensor. The attention weight matrix is ​​then applied to the potential stacking tensor using a dot-multiplication weighting method to form a three-dimensional wiring path potential tensor.

2. The machine learning-based integrated power supply box internal wiring optimization method according to claim 1, characterized in that, Step one: Perform three-dimensional spatial modeling of the electronic components, terminal interfaces, and connection requirements within the integrated power supply box, and construct a spatial topology diagram, specifically as follows: Each electronic component in the integrated power supply box is modeled in three dimensions according to its physical size, mounting plane, and mounting orientation, and a unique spatial coordinate system origin and orientation vector are assigned to each electronic component. The position coordinates, wiring direction, connector type, pin arrangement and corresponding signal type of each terminal interface are structured and encoded, and a local spatial constraint relationship between the terminal interface and electronic device is established based on the terminal interface. Analyze the electrical connection requirements between various electronic devices, and construct directed connection edges based on the starting node and the target node in the connection pair; Based on the electronic device nodes, terminal interface nodes, and connection edges completed in 3D space modeling, construct a spatial topology diagram.

3. The machine learning-based integrated power supply box internal wiring optimization method according to claim 1, characterized in that, Step three: Based on the three-dimensional wiring path potential tensor, a wiring path set is initially generated, and the routing trajectory, electrical parameters, and wiring sequence of each wiring path are recorded to form a wiring path history record set, specifically: For each potential channel corresponding to a wiring connection pair in the three-dimensional wiring path potential tensor, a greedy search strategy is used to extract the path starting from the starting node of the connection pair. The greedy search strategy includes calculating the potential value of adjacent grid points in the three-dimensional spatial grid, starting from the location of the starting node, and selecting the grid point with the largest potential value as the next routing position in the current step. The potential value is the probability score of the spatial grid point corresponding to the starting node to the target node of the current routing connection pair in the three-dimensional routing path potential tensor. The probability score is determined by the diffusion transformer propagating the node features through multiple discrete-time diffusion steps and dynamically weighting and updating it through the attention mechanism, which represents the relative strength of the feasibility of the routing channel at the current spatial location point. Mark the currently selected grid point as a visited node, and proceed step by step until the target node of the corresponding connection pair is reached, thus completing the search for an initial wiring path; During the path search process, the spatial coordinate sequence of each hop is recorded to form a complete path trajectory; Based on the information of the electronic devices connected to each wiring path, the current level, voltage level and signal frequency carried by the wiring path are extracted and associated to obtain the electrical parameters corresponding to the wiring path; The routing path, electrical parameters, and the generated wiring sequence during the search process are numbered and archived to form a wiring path history record set.

4. The machine learning-based integrated power supply box internal wiring optimization method according to claim 1, characterized in that, The identification of electromagnetic interference sensitive areas specifically includes: Map the routing trajectory of each routing path in the routing path history set to a three-dimensional spatial mesh to construct a spatial routing distribution map; In the three-dimensional spatial grid, the wiring path superposition density, signal frequency overlap, and total current intensity superposition of each spatial grid cell are statistically analyzed, and the minimum spacing between adjacent wiring paths is measured to construct an electromagnetic interference characteristic matrix. Spatial grid cells that meet any of the conditions are marked as electromagnetic interference sensitive areas, and corresponding spatial identification indexes are established: Condition 1: The overlap density of wiring paths exceeds a preset density threshold; Condition 2: More than 3 wiring paths exhibit signal frequency overlap within a spatial range; Condition 3: The minimum spacing between adjacent wiring paths is lower than the manufacturing tolerance threshold; Condition 4: The total intensity of the path current exceeds the preset environmental current limiting threshold.

5. The machine learning-based integrated power supply box internal wiring optimization method according to claim 1, characterized in that, The process of constructing a reverse electromagnetic backtracking disturbance field based on the electromagnetic interference sensitive area and the historical data set of the wiring path, and detecting and extracting a subset of interference paths whose overall disturbance score exceeds a preset upper limit threshold, specifically involves: The three-dimensional routing trajectories and corresponding electrical parameters of each wiring path in the wiring path history record set are mapped to the spatial grid where the electromagnetic interference sensitive area is located, and the path segments that cross any electromagnetic interference sensitive area in all wiring paths are selected as the starting point of disturbance propagation. Define a three-dimensional perturbation score tensor consistent with the wiring space structure to represent the perturbation intensity distribution of each spatial grid cell, and set the initial value to zero; For each path segment that traverses an area sensitive to electromagnetic interference, the grid position where the path segment is located is taken as the starting node, and the path segment is traversed back segment by segment towards the starting node in reverse order of the coordinate index of the path segment in the wiring path history record set. At each path segment, the disturbance contribution value is determined by combining the current intensity of the path segment, the signal frequency, and the hop distance from the electromagnetic interference sensitive area. Specifically, the current intensity of the path segment is multiplied by the signal frequency, and then multiplied by a weighting coefficient that decreases exponentially with the number of hops, so that the larger the number of hops, the smaller the disturbance contribution value. Write the disturbance contribution value of each path segment into the corresponding three-dimensional spatial grid position, and accumulate the existing disturbance contribution value at the three-dimensional spatial grid position until all wiring paths have completed back propagation to obtain the reverse electromagnetic back propagation disturbance field. Based on the routing trajectory of each routing path in the historical routing history set, the disturbance contribution value of the routing path at the corresponding three-dimensional spatial grid position in the reverse electromagnetic backtracking disturbance field is extracted and integrated to form the overall disturbance score of the routing path. Wiring paths whose overall disturbance score exceeds a preset upper limit threshold are selected as a subset of interference paths.

6. The machine learning-based integrated power supply box internal wiring optimization method according to claim 1, characterized in that, The specific steps for performing local rerouting on the subset of interference paths are as follows: The three-dimensional trace of each path in the interference path subset is analyzed segment by segment to determine whether the local disturbance score in the reverse electromagnetic backtracking disturbance field exceeds the preset local interference threshold. If the local disturbance score of the path segment is lower than or equal to the preset local disturbance threshold, the original path remains unchanged. If the local disturbance score of a path segment is higher than the preset local interference threshold, a local rerouting is performed. The local rerouting is to re-perform a local greedy search in the three-dimensional cabling path potential tensor to find a new path segment with the highest potential value that does not cross an electromagnetic interference sensitive area. If found, the corresponding path segment in the original path is replaced with the new path segment to obtain the cabling path correction set. Otherwise, the path segment is marked as a sensitive path segment.

7. The machine learning-based integrated power supply box internal wiring optimization method according to claim 1, characterized in that, Step six: Constructing a virtual impedance model based on the sensitive path segment, simulating the electromagnetic response characteristics of the sensitive path segment at different frequencies, and determining easily coupled points, specifically: The three-dimensional routing structure and corresponding electrical parameters of the sensitive path segment are input into the electromagnetic field simulation environment. Distributed parameters composed of series and parallel combinations of resistors, inductors and capacitors are established between the two end nodes of the sensitive path segment to form a virtual impedance model of the sensitive path segment. The virtual impedance model is used to perform impedance-frequency sweep simulation on the sensitive path segment over a wide frequency range to obtain the impedance-frequency characteristic curves of the sensitive path segment at different frequency points. Based on the impedance-frequency characteristic curve, the location where the impedance changes more than a preset change threshold within a unit frequency range near the target operating frequency point and is prone to electromagnetic coupling with adjacent paths is identified as the easily coupled point of the sensitive path segment.

8. The machine learning-based integrated power supply box internal wiring optimization method according to claim 1, characterized in that, Step seven: Insert virtual impedance nodes at easily coupled points in the sensitive path segment, and enhance the anti-interference capability of the path through local layout adjustments, ultimately forming a set of optimized cabling paths, specifically as follows: Virtual impedance nodes are inserted at the easily coupled points of each sensitive path segment. The virtual impedance nodes are equivalent impedance units, which are composed of inductors, capacitors and resistors connected in series. The equivalent impedance value is based on the minimum impedance value in the frequency range where impedance changes occur near the target operating frequency point in the virtual impedance model. It is used to introduce anti-phase impedance matching at the easily coupled points to attenuate the coupling interference energy propagating along the frequency point. Based on the local remaining empty space in the wiring space and the structural assembly constraints, the position of the virtual impedance node is adjusted by displacement, and the insertion point is selected in the metal shell shielding edge, dielectric insulation gap or low frequency interference zone to avoid adding new coupling hot spots. All sensitive path segments that have undergone anti-interference enhancement processing are merged with the routing path correction set to output the final routing optimized path set.

Images