Real-time monitoring method for substation steel structure tilt considering environmental factors

Abstract

The present application relates to substation structure safety monitoring technical field, specifically relates to substation steel structure inclination real-time monitoring method considering environmental factors, the present application extracts the inclination increment sequence and effective sunshine radiation sequence, constructs the heat consumption cost matrix based on the upper limit of heat consumption time of steel and carries out time sequence alignment by utilizing optimal transmission algorithm, extracts time sequence alignment cost, simultaneously, the transfer matrix is reconstructed into the connected network diagram to calculate the day and night cycle aggregation degree. The combined wind characteristic, time sequence alignment cost and day and night cycle aggregation degree calculate comprehensive subsidence index, and compare with the preset normal fluctuation baseline to determine the real ground subsidence. The present application abandons the strict physical temperature difference model, can effectively identify and contain the normal elastic deformation due to heat lag, and completely filters out the false alarm caused by temperature misplacement.

Description

Technical Field

[0001] This invention relates to the field of substation structural safety monitoring technology, specifically to a real-time monitoring method for the tilt of substation steel structures that takes environmental factors into account. Background Technology

[0002] In substation engineering, large steel structures serve as the foundation supporting ultra-high voltage busbars and transmission and transformation equipment. Due to changes in groundwater levels or geological evolution, substation foundations are at risk of uneven settlement, which could lead to permanent tilting of the structure and potentially trigger serious power grid safety accidents. Currently, on-site engineering projects typically deploy dual-axis tilt sensors and surface temperature sensors on the structure to monitor its attitude changes in real time.

[0003] However, substation structures are exposed to a complex outdoor natural environment for extended periods, and the actual slow settlement signals are masked by two significant environmental disturbances. First, there is wind load interference; strong gusts can cause brief, slight, high-frequency swaying of the structure. Second, there is thermal expansion hysteresis interference; after absorbing solar radiation, large steel structures require a considerable physical time to conduct heat into the structure, meaning that the moment the structure deforms due to thermal expansion always lags significantly behind the actual moment of heating.

[0004] Existing monitoring technologies typically employ a static threshold judgment model, directly comparing the tilt angle and temperature values ​​at the same moment to see if they exceed a fixed threshold. This approach cannot address the hysteresis phenomenon caused by the time-series misalignment of heating and deformation. When the structure experiences significant delayed elastic deformation in the afternoon or evening, the existing technology is highly susceptible to misinterpreting this normal heating deformation as irreversible foundation settlement because the solar radiation temperature has already decreased. Furthermore, existing early warning models can only indicate whether the current tilt angle exceeds the limit, failing to calculate the remaining safety buffer period based on the rate of structural deterioration, resulting in a lack of intuitive expectations for on-site engineers to allocate emergency repair materials. Therefore, there is an urgent need for a monitoring method that can explicitly separate environmental heating hysteresis and gust oscillations over time and qualitatively differentiate actual foundation settlement. Summary of the Invention

[0005] To address the technical problem of misjudging actual foundation settlement due to the inability of existing monitoring technologies to handle thermal hysteresis caused by static threshold comparison, the present invention aims to provide a real-time monitoring method for the tilt of substation steel structures that considers environmental factors. The specific technical solution adopted is as follows: A method for real-time monitoring of the tilt of substation steel structures considering environmental factors, the method comprising: The real-time tilt angle and surface temperature of the substation steel frame are obtained, and the tilt increment sequence and gust characteristics are extracted. The direction weight is determined based on the solar azimuth and the orientation of the substation steel frame. The effective solar radiation sequence is generated using the direction weight and the surface temperature. A heat transfer time cost matrix is ​​constructed based on the preset upper limit of heat transfer time for steel; the tilted incremental sequence, the effective solar radiation sequence, and the heat transfer time cost matrix are used as inputs to the optimal transfer algorithm, and a transition matrix is ​​output; the time alignment cost is determined based on the transition matrix; the transition matrix is ​​reconstructed into a connected network graph, and the day-night cycle clustering degree of the connected network graph is calculated and extracted. By combining the gust characteristics, the time alignment cost, and the day-night cycle aggregation degree, a comprehensive subsidence index is calculated, and the comprehensive subsidence index is compared with a preset normal fluctuation baseline to determine whether the substation steel structure has experienced actual foundation subsidence.

[0006] Furthermore, the extraction of the tilt increment sequence and gust features includes: The real-time tilt angle is low-pass filtered to separate the high-frequency oscillation component and the low-frequency tilt angle component; the proportion of the integral of the absolute value of the high-frequency oscillation component within the current preset sampling step size to the integral of the absolute value of the real-time tilt angle before filtering is calculated as the gust characteristic. The low-frequency tilt component is sampled and extracted according to a preset sampling step size, loaded into a sliding time window, and the low-frequency tilt component within the sliding time window is subjected to first-order difference to generate the tilt increment sequence.

[0007] Further, determining the direction weights includes: Obtain the solar azimuth angle at the current time node and the horizontal offset angle of the substation steel structure; calculate the angle between the solar azimuth angle and the horizontal offset angle, and use the absolute value of the cosine of the angle as the direction weight.

[0008] Further, the generation of the effective solar radiation sequence includes: Subtract the lowest temperature value within the sliding time window from the surface temperature within that time window to extract a non-negative temperature rise data sequence; multiply each data point in the temperature rise data sequence by the directional weight of the corresponding time node to determine the effective solar radiation sequence.

[0009] Furthermore, the construction of the heat transfer time cost matrix includes: Each time node in the tilted incremental sequence corresponds to the time when deformation occurs, and each time node in the effective solar radiation sequence corresponds to the time when heat input occurs; based on the chronological relationship of each time node and the upper limit of heat transfer time, a heat transfer time cost matrix is ​​constructed: When the deformation occurs earlier than the heat input time, a blocking constant is configured for the corresponding heat transfer time cost matrix element. When the deformation occurs no earlier than the heat input time and the time difference between the deformation and the heat input time is less than or equal to the upper limit of heat transfer time, the basic constant is configured for the corresponding heat transfer time cost matrix element. If the deformation occurs no earlier than the heating input time and the time difference between the deformation and the heating input time is greater than the upper limit of the heat transfer time, then an exponentially increasing penalty function is configured for the corresponding heat transfer time cost matrix element.

[0010] Furthermore, the method for obtaining the timing alignment cost includes: The effective solar radiation sequence and the tilt increment sequence were normalized respectively. The normalized effective solar radiation sequence, the normalized tilt increment sequence, and the heat transfer time cost matrix are input into the optimal transfer algorithm for iterative solution to obtain the optimal transfer plan matrix as the transfer matrix. The transition matrix is ​​multiplied element-wise with the heat transfer time cost matrix and then summed globally to obtain the time alignment cost.

[0011] Furthermore, the method for obtaining the day-night cycle aggregation degree includes: Add the transition matrix to the transpose of the transition matrix to generate a symmetric connected matrix; The connectivity matrix is ​​sparsified to generate the connectivity network graph. Each time node in the connectivity network graph is regarded as a topological node, and the matrix value is regarded as the weight of the connected edge. The community detection algorithm is used to calculate the connectivity network graph, and the obtained score is used as the day-night cycle clustering degree.

[0012] Furthermore, the calculation of the comprehensive subsidence index includes: The topological dispersion factor is obtained by subtracting the day-night cycle aggregation degree from the preset constant; the normalized time-series alignment cost is multiplied by the topological dispersion factor to generate the numerator; the smoothing term constructed based on the gust characteristics is used as the denominator; the numerator is divided by the denominator to obtain the comprehensive sinking index.

[0013] Furthermore, the determination of whether the substation steel structure has experienced actual foundation settlement includes: If the current comprehensive subsidence index is greater than the preset normal fluctuation baseline, it is determined that the substation steel structure has experienced actual foundation subsidence; if the current comprehensive subsidence index is less than or equal to the preset normal fluctuation baseline, it is determined that no actual foundation subsidence has occurred.

[0014] Furthermore, after determining that the substation steel structure has experienced actual foundation settlement, the process also includes: If it is determined that the substation steel structure has experienced actual foundation settlement, the tilt increment corresponding to the latest sliding time node at the end of each sliding time window determined to be actual foundation settlement is captured, and the extracted tilt increments are accumulated to generate the cumulative residual tilt angle. The cumulative residual tilt angles saved at multiple historical moments are obtained to form a historical tilt angle sequence; Perform a first-order linear regression operation on the historical tilt angle sequence to obtain the tilt evolution rate; if the tilt evolution rate is greater than zero, divide the difference between the preset standard tilt tolerance limit and the current cumulative residual tilt angle by the tilt evolution rate to obtain the remaining step size; multiply the remaining step size by the time conversion factor to convert it into the number of days remaining to reach the danger limit and issue an alarm.

[0015] The present invention has the following beneficial effects: This invention acquires the real-time tilt angle and surface temperature of the steel structure of a substation, extracts the tilt increment sequence and gust characteristics, and weights the effective solar radiation sequence based on the angle between the solar azimuth and the sensor. Addressing the problem of frequent false alarms caused by the inability of existing technologies to handle heating hysteresis due to static threshold comparison, this invention constructs a heat transfer time cost matrix based on the upper limit of steel heat transfer time, uses an optimal transmission algorithm to temporally align the tilt increment and effective solar radiation, and extracts the temporal alignment cost. Simultaneously, it symmetrically transforms the transfer matrix into an undirected connected network graph and extracts the diurnal cycle aggregation degree to quantify the deformation cycle pattern. This invention combines gust characteristics, temporal alignment cost, and diurnal cycle aggregation degree to calculate a comprehensive subsidence index. By comparing this index with a preset normal fluctuation baseline, it determines whether actual foundation subsidence has occurred, thereby effectively identifying and encompassing normal elastic deformation and completely filtering out false alarms. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for real-time monitoring of the tilt of a substation steel structure, taking into account environmental factors, according to an embodiment of the present invention. Detailed Implementation

[0017] The following description, in conjunction with the accompanying drawings, details a specific scheme for a real-time monitoring method for the tilt of a substation steel structure that takes environmental factors into account, provided by the present invention.

[0018] Please see Figure 1 The diagram illustrates a method flowchart for real-time monitoring of substation steel structure tilt considering environmental factors, according to an embodiment of the present invention. The method includes: Step S101: Obtain the real-time tilt angle and surface temperature of the substation steel frame, extract the tilt increment sequence and gust characteristics; determine the direction weight based on the solar azimuth and the orientation of the substation steel frame, and generate an effective solar radiation sequence using the direction weight and the surface temperature.

[0019] In the outdoor environment of substations, the actual slow settlement signal of the steel structure is significantly masked by environmental interference. Therefore, it is necessary to acquire real-time tilt angle and surface temperature to capture the dynamic changes of the structure, i.e., to acquire the real-time tilt angle and surface temperature of the substation steel structure. Specifically, dual-axis tilt angle sensors and surface temperature sensors are deployed at key stress or easily deformable nodes of the target steel structure in the substation. The raw tilt angle data and surface temperature data of the structure are continuously and synchronously acquired through a data acquisition terminal at a preset high-frequency sampling rate. The raw tilt angle data of the dual axes is the real-time tilt angle, including the real-time tilt angle of the X-axis and the real-time tilt angle of the Y-axis, which are orthogonal to each other. A unified timestamp is added to each set of acquired data to obtain the real-time tilt angle data stream and surface temperature data stream containing the timestamp, which are used as the real-time tilt angle and surface temperature of the substation steel structure.

[0020] It should be noted that the preset high-frequency sampling rate is set to be greater than or equal to twice the highest wind-induced vibration frequency of the structure; in this embodiment, it is preferably 1Hz. This setting is used to ensure complete capture of the transient high-frequency swaying characteristics caused by gusts and to provide raw data support that satisfies the Nyquist sampling law for the subsequent low-pass filter to separate the low-frequency tilt component.

[0021] By extracting gust characteristics from the real-time tilt angle of the substation steel structure, the intensity of wind interference can be quantified, thus avoiding false alarms during strong gusts. Processing the real-time tilt angle to generate a tilt increment sequence can strip away long-term background shifts, highlighting the actual new deformation within a unit step. Furthermore, since the three-dimensional structure of the structure is directionally sensitive to solar radiation, a single temperature scalar cannot reflect the actual lateral heating intensity. Therefore, it is necessary to determine directional weights based on the solar azimuth and sensor orientation, and use these weights for weighted mapping to generate an effective solar radiation sequence with spatial directional sensitivity.

[0022] In some possible implementations of this invention, the extraction of the tilt increment sequence and gust features includes: First, the real-time tilt angle is low-pass filtered to separate the high-frequency oscillation component and the low-frequency tilt angle component. Specifically, a preset low-pass filter is used to filter the real-time tilt angle of the X-axis and the real-time tilt angle of the Y-axis respectively, dividing the two axis signals into a high-frequency oscillation component greater than or equal to the cutoff frequency and a low-frequency tilt angle component less than the cutoff frequency.

[0023] As a specific example, the preset low-pass filter can be a Butterworth low-pass filter, and its cutoff frequency is preferably set to 0.01Hz. This filter has the flattest frequency response characteristics within the passband, and its 0.01Hz cutoff frequency setting can effectively isolate short-duration high-frequency swaying caused by gusts, while retaining without distortion the slow tilt angle change trend of long period and extremely low frequency caused by thermal expansion or foundation subsidence, thereby accurately achieving the initial physical separation of wind load interference from real slow deformation signals.

[0024] Further, the proportion of the integral of the absolute value of the high-frequency oscillation component within the current preset sampling step size to the integral of the absolute value of the real-time tilt angle before filtering is calculated as the gust characteristic. More specifically: the integral value of the absolute value of the high-frequency oscillation component within the current preset sampling step size is calculated as the numerator, and the integral value of the absolute value of the real-time tilt angle before filtering within the corresponding time period is calculated as the denominator. To prevent program crashes due to a zero denominator in a completely windless and static state, a preset minimal positive number is appended to the denominator. The ratio of the numerator to the denominator after appending the minimal positive number is calculated as the gust characteristic. This gust characteristic characterizes the proportion of short-duration vibrations caused by the current strong wind in the overall tilt signal.

[0025] As a specific example, the preset sampling step size is set to 15 minutes. This duration accurately captures the low-frequency hysteresis deformation trend caused by changes in ambient temperature while effectively reducing the computational load. Combined with a preset sliding time window containing 96 nodes, the preset sliding time window spans exactly 24 hours, meaning it can hold 96 data points, covering a complete day-night cycle. This provides a complete timescale basis for accurately extracting and measuring the day-night cycle clustering.

[0026] As a specific example, the preset minimum positive number is set to 10. -6 The preset minimum positive number is used as a parameter adjustment factor, and its value is selected to be much smaller than the integral value of the absolute value of the real-time tilt angle during normal operation. This setting ensures that the division operation does not result in division-to-zero deadlock, and also does not cause substantial numerical deviation interference to the calculation results of gust characteristics under normal wind conditions.

[0027] Finally, according to the preset sampling step size, the separated X-axis low-frequency tilt components and Y-axis low-frequency tilt components are sampled and extracted at fixed intervals, and the extracted node values ​​are sequentially loaded into corresponding sliding time windows of preset length to form X-axis low-frequency tilt sequences and Y-axis low-frequency tilt sequences. Subsequently, a first-order difference operation is performed in parallel on the X-axis low-frequency tilt sequences and Y-axis low-frequency tilt sequences within the sliding time windows to generate X-axis tilt increment sequences and Y-axis tilt increment sequences, which together serve as the tilt increment sequences.

[0028] To transform static degrees containing historical absolute offsets into purely new deformations, the N time nodes within the current sliding time window are differentially analyzed sequentially. Specifically, for the first time node within the current sliding time window, the tilt angle value of the preceding historical node on the timeline that is immediately before this first time node and has just exited the time window is extracted as the initial comparison benchmark, and the absolute value of the tilt angle value of the first time node is obtained by subtracting the initial comparison benchmark; for the subsequent time nodes from the 2nd to the Nth time nodes within the current sliding time window, the tilt angle value of the current time node is subtracted from the tilt angle value of its preceding adjacent time node on the timeline, and the absolute value is obtained.

[0029] Through the above complete traversal difference operation, the present invention finally generates an X-axis tilt increment sequence and a Y-axis tilt increment sequence composed of N absolute increment values. These two sequences accurately depict the real new smooth deformation action that occurs between each adjacent sampling step in the dual-axis direction of the substation steel frame.

[0030] In some possible implementations of this invention, determining the direction weight includes: First, the solar azimuth angle at the current time point and the horizontal offset angle of the substation steel structure are obtained. The horizontal offset angle of the substation steel structure includes the X-axis offset angle and the Y-axis offset angle. Since the substation steel structure is a three-dimensional structure, a single ambient temperature scalar cannot reflect the actual amount of solar radiation absorbed on a specific side; therefore, spatial orientation features need to be introduced. The timestamp information of the current time point is extracted, and the pre-configured latitude and longitude parameters of the substation location are retrieved and input into a preset solar calendar calculation model to calculate the solar azimuth angle at the current moment. Simultaneously, the X-axis and Y-axis offset angles of the steel structure relative to true north, calibrated during the installation of the on-site dual-axis tilt sensors, are retrieved.

[0031] It should be noted that the pre-configured latitude and longitude parameters of the substation location are absolute geographical coordinates that are manually entered and fixed in the local register during initial deployment, and are used to provide a unique spatial reference for calculating the sun's position; the preset solar calendar calculation model is a built-in conventional astronomical orbit algorithm. This model uses the latitude and longitude parameters and timestamp information as input parameters, and can calculate and output the sun's azimuth angle at a specific location on Earth at the corresponding time in real time.

[0032] Next, the angle between the solar azimuth angle and the horizontal offset angle is calculated, and the absolute value of the cosine of the angle is used as the direction weight. The angles between the solar azimuth angle and the X-axis horizontal offset angle and the Y-axis horizontal offset angle are calculated respectively. Subsequently, the cosine values ​​of the two angles are calculated respectively. To avoid the logical reversal of the positive and negative signs of subsequent weights due to negative cosine values, the absolute values ​​of these two cosine values ​​are taken to generate the X-axis direction weight and Y-axis direction weight respectively. The cosine value reflects the magnitude of the normal projection component of sunlight on a specific side of the steel structure. The larger the projection component, the stronger the direct thermal radiation absorbed by that side. Through the above processing, the present invention successfully transforms non-directional temperature changes into environmental heating characteristics with spatial direction sensitivity.

[0033] In some possible implementations of this invention, generating an effective solar radiation sequence includes: First, a sliding window is applied to the surface temperature sequence. The lowest temperature value within each sliding time window is subtracted from the current surface temperature within that window to extract a non-negative temperature rise data sequence. More specifically, the lowest temperature value within each sliding time window is searched iteratively. Then, this lowest temperature value is subtracted from each surface temperature data point in the sequence. The extracted temperature rise data sequence is strictly non-negative under any natural environment. This non-negativity constraint is a crucial prerequisite for ensuring that the probability distribution quality matrix in the subsequent optimal transmission algorithm meets mathematical calculation requirements and for avoiding iterative divergence or computational crashes.

[0034] Next, each data point in the temperature rise data sequence is multiplied by the directional weight at the corresponding time node to generate the effective solar radiation sequence. This effective solar radiation sequence includes an X-axis effective solar radiation sequence and a Y-axis effective solar radiation sequence. Each data point in the temperature rise data sequence is multiplied by the X-axis and Y-axis directional weights at the corresponding time node to generate the X-axis and Y-axis effective solar radiation sequences, respectively. This weighted multiplication combines the magnitude of a single temperature change with the normal projection law of space solar radiation, ultimately generating the X-axis and Y-axis effective solar radiation sequences. The numerical values ​​in this effective solar radiation sequence accurately quantify the actual direct thermal radiation intensity experienced by a specific side of the steel structure at a corresponding time, forming the heat source input benchmark for subsequent analysis of thermal expansion hysteresis phenomena.

[0035] Step S102: Construct a heat transfer time cost matrix based on the preset upper limit of heat transfer time for steel; use the tilted incremental sequence, effective solar radiation sequence, and the heat transfer time cost matrix as inputs to the optimal transfer algorithm, and output a transition matrix; determine the time alignment cost based on the transition matrix; reconstruct the transition matrix into a connected network graph, and calculate and extract the day-night cycle clustering degree of the connected network graph.

[0036] Because the heat transfer from steel to the interior of the structure after absorbing solar radiation takes a considerable amount of time, the moment when the structure deforms due to thermal expansion lags significantly behind the actual time of heating. Therefore, it is necessary to set an upper limit for the heat transfer time to define the reasonable time difference between the heat input and the occurrence of deformation. By constructing a heat transfer time cost matrix and assigning values ​​to the corresponding elements in the matrix, the algorithm can be forced to exclude these unreasonable cases when searching for the best alignment scheme, thereby ensuring that subsequent analysis is based on physical laws.

[0037] Using the tilt increment sequence, effective solar radiation sequence, and the heat transfer time cost matrix as inputs to the optimal transfer algorithm is to quantify whether the current tilt increment belongs to normal thermal hysteresis deformation. The tilt increment sequence represents the numerical distribution of the actual gradual deformation action of the structure, the effective solar radiation sequence represents the directional heating characteristics, and the cost matrix represents the time alignment penalty rule. After all three are input, the optimal transfer algorithm can find an allocation scheme with the lowest global transport cost, that is, the optimal alignment ratio with the minimum time penalty, and thus output an optimal transfer matrix.

[0038] The output transition matrix reflects the optimal mapping relationship between heating and deformation within a reasonable lag time. The temporal alignment cost is obtained through this transition matrix. The larger the temporal alignment cost, the more difficult it is to find a sufficient heating cause for the current deformation within a reasonable lag time. In this case, the tilt is likely caused by actual external foundation subsidence.

[0039] The optimal transfer plan matrix output by the optimal transfer algorithm is symmetrically reconstructed into a connected network graph, and the diurnal cycle clustering degree of this connected network graph is extracted to measure the degree of disorder and collapse of the inherent periodic regularity of heating and deformation in the diurnal cycle. In a normal diurnal cycle, regular heating and deformation cause the non-zero elements in the transfer plan matrix to exhibit a concentrated, blocky clustering state, corresponding to a high diurnal cycle clustering degree. However, when the structure experiences sudden structural displacement, the daily periodic regularity of heating and deformation collapses, and the transfer plan matrix exhibits a chaotic characteristic, corresponding to a low diurnal cycle clustering degree. Therefore, the diurnal cycle clustering degree can serve as a quantitative indicator to determine whether the current deformation still follows the daily elastic hysteresis cycle. The diurnal cycle clustering degree and the temporal alignment cost together provide a basis for the generation of the comprehensive subsidence index.

[0040] In some possible implementations of this invention, the construction of the heat transfer time cost matrix includes: Each time node in the tilted incremental sequence corresponds to the time when deformation occurs, and each time node in the effective solar radiation sequence corresponds to the time when heat input occurs; based on the chronological relationship of each time node and the upper limit of heat transfer time, a heat transfer time cost matrix is ​​constructed: When the deformation occurs earlier than the heat input time, a blocking constant is configured for the corresponding heat transfer time cost matrix element. When the deformation occurs no earlier than the heat input time and the time difference between the deformation and the heat input time is less than or equal to the upper limit of heat transfer time, the basic constant is configured for the corresponding heat transfer time cost matrix element. If the deformation occurs no earlier than the heating input time and the time difference between the deformation and the heating input time is greater than the upper limit of the heat transfer time, then an exponentially increasing penalty function is configured for the corresponding heat transfer time cost matrix element.

[0041] When constructing the heat transfer time cost matrix, the upper limit of the heat transfer time required for heat to conduct from the surface of the steel structure to the interior and induce macroscopic expansion is first calculated based on the characteristic thickness of the steel and the thermal diffusivity. This upper limit of the heat transfer time defines the reasonable time difference boundary between the heat input and the occurrence of deformation.

[0042] As a specific example, the formula for calculating the upper limit of the heat transfer time of the preset steel is expressed as follows: .in, Indicates the upper limit of heat transfer time; Indicates the thermal diffusivity of steel; This indicates the characteristic thickness parameter of the steel.

[0043] It should be noted that the thermal diffusivity of steel determines the rate at which temperature homogenizes within the material. It is directly related to the specific material of the steel and can be calculated by dividing the thermal conductivity of that steel type by the product of its density and specific heat capacity, using a table. The characteristic thickness parameter of the steel represents the maximum heat transfer depth path of the main load-bearing component of the substation steel frame, as measured on-site, representing the farthest distance required for heat to penetrate the structure. This physical calculation formula accurately maps the upper limit of the maximum physical time required for heat to conduct from the surface to the deeper layers of the frame and trigger overall macroscopic expansion.

[0044] Subsequently, since the initial data has been decoupled into bidirectional channels of X-axis and Y-axis, independent and parallel matrix construction processes will be executed for the X-axis data and Y-axis data respectively.

[0045] It should be noted that the total number of nodes N corresponds to the length of the sequence being analyzed. The total number of nodes N within the sliding time window is calculated by dividing the preset sliding time window by the preset sampling step size; as a specific example, in this embodiment of the invention, the total number of nodes is 24 hours divided by 15 minutes, resulting in a total number of nodes of 96.

[0046] During operation, both the effective solar radiation sequence and the tilted increment sequence are synchronously extracted and derived from the same sliding time window with a unified timestamp. This synchronous extraction mechanism based on the same time axis naturally ensures that the two sequences are absolutely consistent in terms of time span and total number of nodes N, thus providing a basic data structure guarantee for constructing a heat transfer time cost matrix with equal dimensions.

[0047] Taking the X-axis as an example, for a sliding time window with N nodes, an N×N heat transfer time cost matrix corresponding to the X-axis is constructed. In this heat transfer time cost matrix, the row index corresponds to the heating input time of the effective solar radiation sequence on the X-axis, and the column index corresponds to the deformation occurrence time of the tilt increment sequence on the X-axis. By traversing all row and column intersections, a two-dimensional mapping relationship between the heated nodes and the deformed nodes is established, and the cost value is filled into each matrix element according to the following three judgment logics based on the chronological relationship of the times: When deformation occurs before the heat input, it violates thermodynamic causality. A preset blocking constant is assigned to the corresponding matrix element. As a concrete example, the preset blocking constant is set to 10. 6 This value is orders of magnitude greater than the time penalty cost under normal alignment. By imposing an extremely high penalty cost, the algorithm is forced to prevent the matching of already occurred deformations to yet-to-occur heated inputs during the iterative optimization process of the subsequent optimal transmission algorithm, thereby ensuring that the timing mapping strictly follows the positive logic of time evolution.

[0048] When the deformation occurs no earlier than the heat input time, and the time difference between the two is less than or equal to the upper limit of heat transfer time, it means that the hysteresis time of the deformation fully conforms to the normal physical hysteresis law of heat conduction. A preset basic constant is configured for the corresponding matrix elements. As a specific example, the basic constant is set to 0. This value setting indicates that within the reasonable range allowed by the physical heat transfer time, the structural deformation that occurs is considered a normal elastic response directly caused by the corresponding heat input. Therefore, no mathematical algebraic penalty is imposed on this thermodynamically consistent timing alignment, allowing the optimal transfer algorithm to prioritize lossless matching of the heat source characteristics and deformation characteristics within this range.

[0049] When the deformation occurs no earlier than the heat input time, but the time difference between the two exceeds the upper limit of the heat transfer time, it means that the lag time of the deformation exceeds the limit of physical conduction. The later the deformation occurs, the lower the probability that it is caused by historical heat. Assign the corresponding exponentially increasing penalty function to the corresponding matrix elements.

[0050] The time difference exceeding the upper limit of the heat transfer time is substituted into a preset exponentially increasing penalty function to calculate the exponentially increasing penalty value that increases sharply with the time difference. As a concrete example, the formula for the exponentially increasing penalty function is expressed as: .in, This represents an exponentially increasing penalty function; This represents the time difference between the moment deformation occurs and the moment of heat input; Indicates the upper limit of heat transfer time; This represents the preset time penalty scaling factor; This represents the natural constant. It should be noted that the preset time penalty scaling factor is set to 0.01; this value ensures that the exponential penalty curve has an effective gradient while preventing the penalty value from increasing too rapidly with the time difference.

[0051] Through the above construction method, the cost matrix successfully transforms the physical boundary of the heat transfer hysteresis of steel into a rigorous mathematical penalty rule, enabling subsequent algorithms to find the best alignment interpretation scheme within a reasonable time delay constraint.

[0052] In some possible implementations of this invention, the method for obtaining the timing alignment cost includes: First, the effective solar radiation sequence and the tilt increment sequence are normalized separately. The optimal transport algorithm, mathematically, mandates that the input alignment sequence must be a probability distribution sequence with a sum of elements strictly equal to 1. The following probability distribution normalization operations are performed independently and in parallel for the effective solar radiation sequence and tilt increment sequence corresponding to the X-axis and the effective solar radiation sequence and tilt increment sequence corresponding to the Y-axis: Taking the X-axis calculation branch as an example, this invention calculates the sum of the values ​​of all elements in the effective solar radiation sequence along the X-axis. To prevent program crashes caused by division by zero under conditions such as constant temperature or nighttime when the heat input approaches zero, this invention adds a preset, extremely small positive number as a zero-prevention denominator to the sum. Subsequently, this invention divides the value of each node in the effective solar radiation sequence along the X-axis one by one by this zero-prevention denominator. Through this calculation, this invention obtains a normalized effective solar radiation sequence whose sum is strictly approximately 1, which is transformed into a probability distribution representing the weighted proportion of heat input within a time window.

[0053] Through the above independent step-by-step processing, the present invention successfully transforms the heating input and deformation increment with different physical dimensions into a unified marginal probability distribution sequence that meets the stringent input conditions of the optimal transmission algorithm.

[0054] Secondly, the normalized X-axis effective solar radiation sequence, the normalized X-axis tilt increment sequence, and the corresponding X-axis heat transfer time cost matrix are input into the optimal transfer algorithm for iterative solution to obtain the optimal transfer plan matrix as the X-axis transfer matrix. Internally, the algorithm continuously scales and iteratively evaluates the total penalty under different time-series pairing methods to find the allocation scheme with the lowest global time cost, and converges to output the optimal X-axis transfer plan matrix as the X-axis transfer matrix. The elements in this matrix represent the optimal time-series mapping ratio between historical heat gain and current deformation along the X-axis. Parallel, the normalized sequence and cost matrix corresponding to the Y-axis are input into the algorithm, and the Y-axis transfer matrix is ​​solved independently. It should be noted that the transfer matrix includes both the X-axis transfer matrix and the Y-axis transfer matrix. It should also be noted that the optimal transfer algorithm described in this embodiment uses the classic Sinkhorn iterative algorithm; the Sinkhorn iterative algorithm is well-known to those skilled in the art and will not be elaborated upon here.

[0055] Finally, the X-axis transition matrix is ​​multiplied element-wise and summed globally to obtain the X-axis time-alignment cost. The obtained X-axis transition matrix is ​​then multiplied element-wise with the corresponding X-axis heat transfer time-cost matrix, and all resulting terms are summed globally. To restore the true physical dimensions, this summation is multiplied by the sum of the absolute values ​​of the X-axis tilt increment sequence before normalization, thus calculating the X-axis time-alignment cost.

[0056] Subsequently, using identical logic, dot product, summation, and dimension restoration were performed on the Y-axis transition matrix and cost matrix to calculate the Y-axis temporal alignment cost. These two cost values ​​serve as core scalars, intuitively quantifying the magnitude of the error caused by tilting deformation along their respective axes that cannot be reasonably explained by thermal hysteresis. The larger the cost value, the more the deformation deviates from the normal thermal expansion and contraction law, revealing a higher probability that it originates from actual external foundation subsidence.

[0057] In some possible implementations of this invention, the method for obtaining the day-night cycle aggregation degree includes: First, the transition matrix is ​​added to its transpose to generate a symmetric connected matrix. The X-axis transition matrix from the preceding solution is extracted. Since this matrix represents a one-way temporal mapping from heating to deformation, it is a directed asymmetric matrix and cannot be directly used to measure the characteristics of undirected communities. Therefore, the X-axis transition matrix is ​​transposed to generate its transpose, and the original X-axis transition matrix is ​​added to the transpose matrix at corresponding positions. Through this symmetric addition process, this invention transforms the one-way temporal mapping into a bidirectional interactive X-axis symmetric connected matrix, establishing a mathematical foundation for the subsequent construction of the undirected network graph.

[0058] Secondly, the symmetric connected matrix is ​​sparsified to generate the connected network graph. The transition matrix output by the optimal transport algorithm usually contains a large number of near-zero numerical computational residuals. If the network is directly constructed, a dense fully connected graph without any features will be formed, causing subsequent algorithms to fail. To address this, the present invention performs truncation sparsification on the symmetric connected matrix, that is, forces weak noise elements in the matrix with values ​​less than a preset connection weight threshold to zero, retaining only the core skeleton elements that reflect strong temporal correspondences, thereby generating a high-fidelity X-axis connected network graph.

[0059] As a specific example, the preset connectivity weight threshold is set to 10. -4 This value can effectively filter out the vast majority of meaningless weak connected edges in the matrix, retaining only the core skeleton transition edges that represent a strong and real temporal mapping relationship between heating and deformation, thus ensuring that the subsequent community detection algorithm can clearly identify the clustering topological features of the day-night cycle.

[0060] Finally, each time node in the connected network graph is considered a topological node, and the matrix values ​​are considered as the weights of connected edges. A community detection algorithm is used to calculate the connected network graph, and the obtained score is used as the day-night cycle clustering degree. In the obtained sparse X-axis connected network graph, each time node within the sliding time window is mapped to an independent topological node in the network topology, and the non-zero values ​​retained in the matrix are used as the weights of connected edges connecting the corresponding nodes. This invention calls a preset community detection algorithm to calculate whether the nodes in the network exhibit tight clustering characteristics. In a normal day-night cycle, heating and elastic deformation will exhibit a highly consistent periodic interaction, and the algorithm will output a high modular score; while when the foundation suffers sudden damage, causing the daily periodic regularity to collapse, the algorithm will output a lower score. This invention extracts the score representing the degree of tight clustering of the network as the X-axis day-night cycle clustering degree.

[0061] It should be noted that the community detection algorithm preferably uses the Louvain algorithm; when executing this algorithm, this invention uses the truncated and sparsified connected network graph as the sole input. In this network graph, N time nodes are defined as independent nodes in the network topology, and the non-zero values ​​retained in the matrix are defined as the weights of the connecting edges between the nodes. This algorithm does not require supervised training; internally, it attempts to maximize the modularity of the network by repeatedly merging tightly connected local nodes. After iterative convergence, the algorithm directly outputs a modularity score ranging from 0 to 1 as the final output. This output score is the aforementioned day-night cycle clustering degree, which is common knowledge to those skilled in the art and will not be elaborated upon here.

[0062] After completing the X-axis calculation, this invention employs the exact same logic to sequentially perform transpose addition, truncation sparsification, and community detection algorithm calculations on the previously obtained Y-axis transition matrix, independently generating the Y-axis day-night cycle clustering degree. Thus, by utilizing pure graph theory features, it achieves independent qualitative and quantitative analysis of the daily elastic collapse state of the structure along both axes.

[0063] Step S103: Combine the gust characteristics, the time alignment cost, and the day-night cycle aggregation degree to calculate the comprehensive subsidence index, and compare the comprehensive subsidence index with the preset normal fluctuation baseline to determine whether the substation steel structure has experienced actual foundation subsidence.

[0064] Current monitoring technologies typically employ a static threshold determination method that directly compares tilt angle and temperature values ​​at the same moment. This approach cannot address the hysteresis caused by the time-series misalignment of heating and deformation. When a substation's steel structure experiences significant delayed elastic deformation in the afternoon or evening, the existing technology is highly susceptible to misinterpreting this normal thermal deformation as irreversible foundation subsidence because the solar radiation temperature has already decreased. Furthermore, existing early warning systems can only indicate whether the current tilt angle exceeds the limit; they cannot calculate the remaining safety buffer period based on the rate of structural deterioration, resulting in on-site engineers lacking a clear expectation for allocating emergency repair materials.

[0065] To address the aforementioned issues, this invention, after completing preliminary signal processing and temporal alignment feature extraction, comprehensively determines whether actual foundation subsidence has occurred by combining gust characteristics, temporal alignment cost, and diurnal cycle clustering degree. Specifically, the temporal alignment cost reflects the degree to which current deformation cannot be explained by historical heating, the diurnal cycle clustering degree reflects whether the daily cyclical pattern of heating deformation has been disrupted, and gust characteristics are used to suppress false subsidence misjudgments caused by strong winds. By jointly calculating a comprehensive subsidence index based on these three factors and comparing it with a preset normal fluctuation baseline, the invention effectively separates environmental elastic deformation from actual foundation subsidence in a temporal manner, completely filtering out false alarms caused by temperature misalignment.

[0066] In some possible implementations of this invention, the calculation of the comprehensive subsidence index includes: The topological dispersion factor is obtained by subtracting the day-night cycle aggregation degree from the preset constant; the normalized time-series alignment cost is multiplied by the topological dispersion factor to generate the numerator; the smoothing term constructed based on the gust characteristics is used as the denominator; the numerator is divided by the denominator to obtain the comprehensive sinking index.

[0067] Since the invention has already extracted the independent temporal alignment cost and day-night cycle clustering degree for both axes, this invention will perform the same exponential calculation logic for both the X and Y axes in parallel. The following steps will be explained using the X-axis calculation process as an example: First, the topological dispersion factor is obtained by subtracting the day-night cycle clustering degree from a preset constant. This invention calculates the topological dispersion factor by subtracting the previously extracted day-night cycle clustering degree from a preset constant. This factor constitutes a reverse penalty mechanism: the lower the clustering degree, the more dispersed the cycle, and the larger the factor value. As a specific example, the preset constant is set to 1 because the theoretical range of the day-night cycle clustering degree is [0,1]. Subtracting the day-night cycle clustering degree from the preset constant of 1 accurately reverses the day-night cycle clustering degree into a continuous penalty multiplier representing the degree of dispersion.

[0068] Secondly, the system establishes a fixed-length historical time-series alignment cost buffer queue. Each time, the current time-series alignment cost is stored in this queue, and the historical maximum and minimum values ​​in the queue are extracted. This scales the current time-series alignment cost to the [0,1] interval, resulting in a normalized time-series alignment cost. As a specific example, the fixed length is set to 288; this value corresponds to a sliding step size of 288 15-minute intervals over three consecutive days, covering the complete environmental day-night cycle and ensuring that the extracted historical extreme values ​​are statistically significant. This invention multiplies the normalized time-series alignment cost by the topological scattering factor to generate a numerator representing the degree of subsidence deterioration.

[0069] Meanwhile, the present invention retrieves the gust features extracted in the preprocessing stage, adds the gust features to the constant 1, and then superimposes a smoothing constant to construct a smoothing term that suppresses false alarms of strong winds and strictly prevents overflow from division by zero, and uses it as the denominator term.

[0070] Finally, the local subsidence index is calculated and aggregated to generate the final comprehensive subsidence index. This invention divides the calculated numerator by the denominator to obtain the uniaxial local subsidence index.

[0071] As a concrete example, taking the X-axis as an example, the local sinking index of the X-axis The calculation formula is expressed as: .in, This represents the time alignment cost of the normalized X-axis; This represents the day-night cycle concentration on the X-axis; Represents the topological scrambling factor; Describe the characteristics of gusts; It is a smoothing constant; This indicates the smoothing term.

[0072] As a concrete example, the smoothing constant is set to 10. -4 Used to prevent computational overflow and ensure numerical stability.

[0073] This formula exhibits a clear multidimensional qualitative pattern: topological dispersion factor. A reversal penalty for abnormal cycles is implemented; the lower the clustering degree, the larger the multiplier term. The division denominator in the formula constructs a wind disturbance smoothing mechanism, which is applied when gust characteristics... At its maximum, the overall index is forcibly suppressed, thus forcibly separating the slight oscillations caused by strong winds. This localized subsidence index... High-risk values ​​will only be output under harsh conditions such as large timing alignment errors, scattered topological patterns, and weak wind interference.

[0074] Because uneven settlement of the substation steel structure foundation often exhibits strong directionality, such as the collapse of a single-sided abutment, dangerous deformation may only manifest as abnormally severe deformation in a single axis, while remaining relatively stable in another axis. If the average value of the two axes is taken, the danger signal from the severe axis will be diluted and masked by the stable axis; if an alarm is only triggered when both axes exceed their limits simultaneously, it is highly likely to result in serious missed alarms.

[0075] Therefore, after calculating the local subsidence index along the X-axis and the local subsidence index along the Y-axis in parallel, the maximum value of these two values ​​is extracted as the comprehensive subsidence index. This maximum value accurately identifies the most severe and dangerous deformation direction of the current structure, ensuring that any real settlement in any single dimension can trigger the subsequent anomaly interception module without damage, thereby achieving a high-confidence qualitative judgment on the potential for real foundation subsidence.

[0076] In some possible implementations of this invention, determining whether the substation steel structure has experienced actual foundation settlement includes: If the current comprehensive subsidence index is greater than the preset normal fluctuation baseline, it is determined that the substation steel structure has experienced actual foundation subsidence; if the current comprehensive subsidence index is less than or equal to the preset normal fluctuation baseline, it is determined that no actual foundation subsidence has occurred.

[0077] First, a localized baseline for normal fluctuations is established. Due to differences in the geographical location, steel frame type, and sensor installation deviations of different substations, this invention requires a pre-installed cold-start calibration mechanism. During the initial stable operation period after the monitoring equipment is installed, this invention does not trigger any early warning actions, but continuously collects and calculates a series of historical comprehensive subsidence indices. It should be noted that, to ensure the calibration data covers sufficient environmental cycle characteristics, this calibration operation period needs to be set to include at least three consecutive complete day-night cycles.

[0078] Once the calibration time requirement is met, this invention calculates the maximum peak value of all comprehensive subsidence index sequences within that period, multiplies the peak value by a preset safety tolerance coefficient, and finally establishes it as the normal fluctuation baseline reflecting the current local health status of the steel structure, and stores it in the local register.

[0079] As a specific example, the preset safety tolerance coefficient can be set to 1.3. This value provides an appropriate tolerance boundary for the present invention, which can prevent the present invention from being too sensitive and frequently generating false alarms, while ensuring that the baseline margin is not too wide, avoiding the underreporting of real and minor settlement in the early stages of the substation structure; the implementer can make adaptive adjustments to the specific value according to the accuracy requirements of the equipment.

[0080] Secondly, during subsequent formal monitoring, each time the sliding time window advances, this invention acquires the current comprehensive subsidence index and performs a rigorous numerical comparison with a fixed, preset normal fluctuation baseline. If the comparison result shows that the current comprehensive subsidence index is less than or equal to the preset normal fluctuation baseline, it indicates that the current minor deformation is entirely within the hysteresis law of environmental thermal expansion or falls within the range of wind noise. In this case, this invention determines that no actual foundation subsidence has occurred. At this point, this invention automatically cuts off the data transmission stream, directly discards the characteristic values ​​within the current time window, and does not perform any permanent offset accumulation and storage.

[0081] If the comparison results show that the current comprehensive subsidence index is significantly greater than the preset normal fluctuation baseline, it indicates that the current tilt state has exceeded the safety limit that can be explained by environmental factors. The present invention will immediately generate an abnormal over-limit trigger command, determine that the substation steel structure is undergoing irreversible real foundation subsidence, and immediately start the subsequent single-point deviation accumulation and business simulation process.

[0082] Once actual foundation settlement is determined, this invention needs to further convert qualitative signals into quantitative data. Since the preceding data stream always slides within a sliding time window containing N nodes, accumulating all increments within the entire window after determination would lead to severe overlapping errors between adjacent sliding windows. Therefore, this invention only captures the X-axis tilt increment and Y-axis tilt increment corresponding to the latest sliding time node at the end of each sliding time window where actual foundation settlement is determined, retaining the original positive and negative signs. Subsequently, the X-axis tilt increment and Y-axis tilt increment with bias polarity captured in each iteration are coaxially accumulated and subtracted to obtain the cumulative residual offset of the X-axis and Y-axis. Finally, the square root of the sum of the squares of the cumulative residual offset of the X-axis and Y-axis is calculated to avoid deadlock due to repeated accumulation of sliding windows, ultimately generating a cumulative residual tilt angle that reflects the true spatial attitude of the structure after the positive and negative rebounds cancel each other out.

[0083] In some possible implementations of this invention, after determining that the substation steel structure has experienced actual foundation settlement, the method further includes: If it is determined that the substation steel structure has experienced actual foundation settlement, the tilt increment corresponding to the latest sliding time node at the end of each sliding time window determined to be actual foundation settlement is captured, and the extracted tilt increments are accumulated to generate the cumulative residual tilt angle. The cumulative residual tilt angles saved at multiple historical moments are obtained to form a historical tilt angle sequence; Perform a first-order linear regression operation on the historical tilt angle sequence to obtain the tilt evolution rate; if the tilt evolution rate is greater than zero, divide the difference between the preset standard tilt tolerance limit and the current cumulative residual tilt angle by the tilt evolution rate to obtain the remaining step size; multiply the remaining step size by the time conversion factor to convert it into the number of days remaining to reach the danger limit and issue an alarm.

[0084] Simply exceeding the instantaneous angle limit can only indicate risk, but cannot provide the maintenance department with a clear repair schedule. Therefore, this invention extracts a set of continuously stored cumulative residual tilt angles from the internally maintained historical tilt register according to a preset regression window length, forming a historical tilt angle sequence. To ensure statistical significance, this sequence contains at least a preset number of valid accumulation nodes.

[0085] It should be noted that the historical tilt register is a dedicated non-volatile storage area designated within this invention. The regression window length defines the time depth traced by this invention when performing linear trend fitting, determining whether the extrapolation result focuses on reflecting recent steep abrupt changes or long-term gradual deterioration trends. To prevent the linear regression algorithm from overfitting due to insufficient effective data points, this invention mandates that the number of effective accumulated nodes within the traced interval must reach a preset threshold. Only when the actual offset data accumulated in the historical tilt register meets this requirement will this invention initiate rigorous statistical regression calculations. As a specific example, the preset threshold is set to 30 effective accumulated nodes. Setting it to 30 effective nodes ensures that the first-order linear regression algorithm, with sufficient data support, smooths out localized random fluctuations and accurately and stably extracts the true subsidence and deterioration trend of the steel structure.

[0086] Subsequently, this invention extracts the time nodes corresponding to the elements in these historical tilt angle sequences when they were saved, sets them as the independent variables on the horizontal axis, and sets the corresponding historical tilt angles as the dependent variables on the vertical axis, forming a two-dimensional mapped sample point series with a real time span; performs a first-order linear regression operation on the two-dimensional mapped sample point series, and extracts the slope value corresponding to the regression fitted line as the tilt evolution rate. This rate is accompanied by a real time dimension, which intuitively represents the steepness of the recent accelerated loss of foundation bearing capacity.

[0087] After extracting the tilt evolution rate, this invention forcibly introduces a mistake-proofing branch: if the tilt evolution rate is less than or equal to zero, it indicates that the structural deviation is in a stagnant or extremely slow stable state. In this case, the invention will stop the calculation and directly output a stable operation log to avoid negative value deadlock or division-by-zero crashes in subsequent division equations. If the tilt evolution rate is determined to be greater than zero, it indicates that the structure is continuously and monotonically deteriorating. In this case, the invention retrieves the preset specification tilt tolerance value corresponding to the target structure from the local database and then performs the calculation and deduction of the remaining days.

[0088] As a concrete example, the formula for calculating the remaining days is expressed as follows: .in, Indicates the number of days remaining; This indicates the preset allowable tilt limit; Indicates the current cumulative residual tilt angle; Indicates the rate of tilted evolution; This indicates the preset sampling step size; 1440 represents the total number of minutes for the entire day.

[0089] In this formula logic, the numerator term This represents the remaining angular margin of the structure from the danger red line; dividing this margin by the evolution rate yields the remaining steps required for the structure to reach the red line. Finally, this invention multiplies this by a time conversion factor. The number of step sizes is successfully mapped to a time period in days that has engineering guidance significance. Through this dimensional conversion, the invention ultimately outputs the remaining days with a clear countdown attribute.

[0090] As a specific example, the lower limit for emergency dispatch is set at 15 days. This lower limit is a reasonable buffer period based on engineering practice experience. When the estimated remaining days fall below this lower limit, it indicates that the structure has entered a high-risk countdown stage before physical collapse. In this case, the invention must immediately trigger an alarm work order to ensure that repair personnel can complete substantial foundation reinforcement intervention before the red line is breached.

Claims

1. A method for real-time monitoring of the tilt of substation steel structures considering environmental factors, characterized in that, The method includes: The real-time tilt angle and surface temperature of the substation steel frame are obtained, and the tilt increment sequence and gust characteristics are extracted. The direction weight is determined based on the solar azimuth and the orientation of the substation steel frame. The effective solar radiation sequence is generated using the direction weight and the surface temperature. A heat transfer time cost matrix is ​​constructed based on the preset upper limit of heat transfer time for steel; the tilted incremental sequence, the effective solar radiation sequence, and the heat transfer time cost matrix are used as inputs to the optimal transfer algorithm, and a transition matrix is ​​output; the time alignment cost is determined based on the transition matrix; the transition matrix is ​​reconstructed into a connected network graph, and the day-night cycle clustering degree of the connected network graph is calculated and extracted. By combining the gust characteristics, the time alignment cost, and the day-night cycle aggregation degree, a comprehensive subsidence index is calculated, and the comprehensive subsidence index is compared with a preset normal fluctuation baseline to determine whether the substation steel structure has experienced actual foundation subsidence. The process of extracting the tilt increment sequence and gust features includes: performing low-pass filtering on the real-time tilt angle to separate the high-frequency oscillation component and the low-frequency tilt angle component; calculating the proportion of the integral of the absolute value of the high-frequency oscillation component within the current preset sampling step size to the integral of the absolute value of the real-time tilt angle before filtering, as the gust feature; sampling and extracting the low-frequency tilt angle component according to the preset sampling step size, loading it into a sliding time window, and performing first-order difference on the low-frequency tilt angle component within the sliding time window to generate the tilt increment sequence. The determination of the direction weight includes: obtaining the solar azimuth angle at the current time node and the horizontal offset angle of the substation steel structure; calculating the angle between the solar azimuth angle and the horizontal offset angle, and using the absolute value of the cosine of the angle as the direction weight; The step of generating an effective solar radiation sequence includes: subtracting the lowest temperature value within the sliding time window from each surface temperature within the sliding time window to extract a non-negative temperature rise data sequence; and multiplying each data point in the temperature rise data sequence by the directional weight of the corresponding time node to determine the effective solar radiation sequence. The construction of the heat transfer time cost matrix includes: each time node in the tilted incremental sequence corresponds to the deformation occurrence time, and each time node in the effective solar radiation sequence corresponds to the heat input time; the heat transfer time cost matrix is ​​constructed according to the time sequence relationship of each time node and the upper limit of heat transfer time. The method for obtaining the time-series alignment cost includes: normalizing the effective solar radiation sequence and the tilted increment sequence respectively; inputting the normalized effective solar radiation sequence, the normalized tilted increment sequence, and the heat transfer time cost matrix into the optimal transfer algorithm for iterative solution to obtain the optimal transfer plan matrix as the transfer matrix; multiplying the corresponding elements of the transfer matrix and the heat transfer time cost matrix and summing them globally to obtain the time-series alignment cost. The method for obtaining the day-night cycle clustering degree includes: adding the transition matrix to the transpose of the transition matrix to generate a symmetric connected matrix; performing sparsification on the connected matrix to generate the connected network graph; treating each time node in the connected network graph as a topological node, the matrix value as the weight of the connected edge, using a community detection algorithm to calculate the connected network graph, and using the obtained score as the day-night cycle clustering degree.

2. The method for real-time monitoring of substation steel structure tilt considering environmental factors according to claim 1, characterized in that, The construction of the heat transfer time cost matrix also includes: When the deformation occurs earlier than the heat input time, a blocking constant is configured for the corresponding heat transfer time cost matrix element. When the deformation occurs no earlier than the heat input time and the time difference between the deformation and the heat input time is less than or equal to the upper limit of heat transfer time, the basic constant is configured for the corresponding heat transfer time cost matrix element. If the deformation occurs no earlier than the heating input time and the time difference between the deformation and the heating input time is greater than the upper limit of the heat transfer time, then an exponentially increasing penalty function is configured for the corresponding heat transfer time cost matrix element.

3. The method for real-time monitoring of substation steel structure tilt considering environmental factors according to claim 1, characterized in that, The calculation of the comprehensive subsidence index includes: The topological dispersion factor is obtained by subtracting the day-night cycle aggregation degree from the preset constant; the normalized time-series alignment cost is multiplied by the topological dispersion factor to generate the numerator; the smoothing term constructed based on the gust characteristics is used as the denominator; the numerator is divided by the denominator to obtain the comprehensive sinking index.

4. The method for real-time monitoring of substation steel structure tilt considering environmental factors according to claim 1, characterized in that, The determination of whether the substation steel structure has experienced actual foundation settlement includes: If the current comprehensive subsidence index is greater than the preset normal fluctuation baseline, it is determined that the substation steel structure has experienced actual foundation subsidence; if the current comprehensive subsidence index is less than or equal to the preset normal fluctuation baseline, it is determined that no actual foundation subsidence has occurred.

5. A method for real-time monitoring of substation steel structure tilt considering environmental factors according to claim 4, characterized in that, After determining that the substation steel structure has experienced actual foundation settlement, the process further includes: If it is determined that the substation steel structure has experienced actual foundation settlement, the tilt increment corresponding to the latest sliding time node at the end of each sliding time window determined to be actual foundation settlement is captured, and the extracted tilt increments are accumulated to generate the cumulative residual tilt angle. The cumulative residual tilt angles saved at multiple historical moments are obtained to form a historical tilt angle sequence; Perform a first-order linear regression operation on the historical tilt angle sequence to obtain the tilt evolution rate; if the tilt evolution rate is greater than zero, divide the difference between the preset standard tilt tolerance limit and the current cumulative residual tilt angle by the tilt evolution rate to obtain the remaining step size; multiply the remaining step size by the time conversion factor to convert it into the number of days remaining to reach the danger limit and issue an alarm.

Images