Steel support axial force sensing data intelligent analysis method for fabricated station

Abstract

The present application relates to the technical field of engineering intelligent monitoring, and especially relates to a steel support axial force sensing data intelligent analysis method for a fabricated station. The method comprises the following steps: collecting an axial force time series data set, an installation position information set and a construction time record set of each steel support in the fabricated station; calculating a baseline axial force reference value of each steel support based on the axial force time series data set; and calculating a spatial attenuation coefficient of each steel support based on the installation position information set and the axial force time series data set and forming a spatial attenuation coefficient data. Therefore, the present application solves the problems of baseline setting relying on experience, spatial difference being difficult to quantify and time effect being discontinuous in the traditional method by establishing a data-driven analysis process based on time series, space and time multi-dimensional characteristics, and improves the scientificity, reliability and operability of steel support axial force monitoring.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology in engineering, and in particular to an intelligent analysis method for axial force sensing data of steel supports in prefabricated railway stations. Background Technology

[0002] Current technologies for monitoring and analyzing axial force in steel supports still have many shortcomings. On the one hand, existing sensor monitoring relies heavily on single-point or small-scale distributed sensor data collection, lacking a unified time benchmark and spatial coordination mechanism. This makes it difficult to directly compare data from different sensor points, resulting in insufficient temporal completeness. On the other hand, the axial force baseline is usually set using manual experience or single static load values ​​as a reference, failing to use statistical methods to identify fluctuations in the initial stage. This easily leads to the accumulation of baseline deviations, affecting subsequent judgments. Spatially, traditional methods do not adequately consider the mutual influence between supports. The stress differences between adjacent components often fail to be quantified as attenuation relationships, making it difficult to accurately characterize the propagation of local anomalies. Temporally, the stress evolution at different stages after construction is complex, but existing methods mostly use fixed reduction coefficients or empirical formulas for approximation, lacking a continuous characterization of the time response process, leading to unclear distinction between short-term and medium-to-long-term effects. Furthermore, in terms of data fusion and anomaly detection, existing technologies often directly compare baseline values ​​with measured values. When deviations occur, it is impossible to clearly distinguish whether they are caused by location factors, time factors, or collaborative factors, which can easily lead to false alarms or missed alarms. There is a lack of a comprehensive judgment mechanism based on multi-dimensional indicators. Finally, in the visualization and alarm stages, most existing systems only output a single numerical value or a simple curve, lacking a traceable data link related to spatial location and temporal evolution, which is detrimental to prioritizing anomaly handling and rapid decision-making. Summary of the Invention

[0003] Therefore, it is necessary to provide an intelligent analysis method for axial force sensing data of steel supports in prefabricated stations to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, an intelligent analysis method for axial force sensing data of steel supports in prefabricated railway stations is provided, the method comprising the following steps:

[0005] Step S1: Collect the axial force time series dataset, installation location information set, and construction time record set for each steel support in the prefabricated station;

[0006] Step S2: Calculate the baseline axial force reference value for each steel support based on the axial force time series dataset;

[0007] Step S3: Calculate the spatial attenuation coefficient of each steel support based on the installation location information set and the axial force time series dataset, and generate spatial attenuation coefficient data;

[0008] Step S4: Calculate the time response coefficient of each steel support based on the construction time record set and the axial force time series dataset, and form a time response coefficient set;

[0009] Step S5: Calculate the baseline axial force reference value, spatial attenuation coefficient data, time response coefficient set and the current observed axial force value according to the predetermined combination rules to obtain the initial axial force estimate value of each steel support and generate the axial force estimate value.

[0010] The beneficial effects of this invention are that it utilizes a distributed sensor network to acquire axial force time-series data of steel supports, and simultaneously records installation location information and construction time, ensuring that each support has a clear spatiotemporal label and guaranteeing the comparability and integrity of the data. By performing windowing processing and statistical judgment on the time-series data, a baseline axial force reference value is obtained, and the stability coefficient is used as a quality criterion, avoiding the limitation of baseline dependence on a single measuring point.

[0011] Secondly, at the level of parameter construction and multi-dimensional fusion, neighborhood relationships were calculated by installing coordinates, and a spatial attenuation coefficient was established by combining baseline values ​​and deviations, thus quantifying the differences in stress within the neighborhood. Simultaneously, a set of time response coefficients was constructed based on the number of days between the construction completion time and the observation time, characterizing the time effect through a combination of piecewise and continuous functions. Furthermore, the baseline deviation, spatial attenuation coefficient, and time response coefficient were fused according to predetermined rules to calculate the comprehensive response index and form an initial value for axial force estimation. This initial value was then paired with measured calibration data to generate training sample pairs, establishing an axial force inversion relationship.

[0012] Finally, at the level of anomaly detection and result presentation, an anomaly index is obtained by retrieving the point-by-point difference between the axial force and the baseline value, and then sorted according to the spatial attenuation coefficient, thereby outputting a priority handling list and visual charts, realizing the intuitiveness of anomaly identification and handling and data traceability. Therefore, this invention solves the problems of baseline setting relying on experience, difficulty in quantifying spatial differences, and discontinuous handling of temporal effects in traditional methods by establishing a data-driven analysis process based on multi-dimensional features of time series, space, and time, thus improving the scientificity, reliability, and operability of axial force monitoring of steel supports. Attached Figure Description

[0013] Figure 1 A flowchart illustrating the steps of an intelligent analysis method for axial force sensing data of steel supports in prefabricated railway stations.

[0014] Figure 2 This is a schematic diagram of the time response coefficient curve;

[0015] Figure 3 A grayscale simulation diagram of the axial force values ​​distributed across the station supports;

[0016] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] The technical method of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0018] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0019] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0020] To achieve the above objectives, please refer to Figures 1 to 3 A method for intelligent analysis of axial force sensing data of steel supports in prefabricated railway stations, the method comprising the following steps:

[0021] Step S1: Collect the axial force time series dataset, installation location information set, and construction time record set for each steel support in the prefabricated station;

[0022] Step S2: Calculate the baseline axial force reference value for each steel support based on the axial force time series dataset;

[0023] Step S3: Calculate the spatial attenuation coefficient of each steel support based on the installation location information set and the axial force time series dataset, and generate spatial attenuation coefficient data;

[0024] Step S4: Calculate the time response coefficient of each steel support based on the construction time record set and the axial force time series dataset, and form a time response coefficient set;

[0025] Step S5: Calculate the baseline axial force reference value, spatial attenuation coefficient data, time response coefficient set and the current observed axial force value according to the predetermined combination rules to obtain the initial axial force estimate value of each steel support and generate the axial force estimate value.

[0026] In this embodiment of the invention, reference Figure 1 The diagram shown is a flowchart illustrating the steps of an intelligent analysis method for axial force sensing data of steel supports in prefabricated railway stations according to the present invention. In this example, the intelligent analysis method for axial force sensing data of steel supports in prefabricated railway stations includes the following steps:

[0027] Step S1: Collect the axial force time series dataset, installation location information set, and construction time record set for each steel support in the prefabricated station;

[0028] Preferably, step S1 includes:

[0029] The distributed strain gauge axial force sensor network is controlled to collect axial force time series data of each steel support under static load and working load according to a unified time baseline. The axial force time series dataset contains an equal time-slot sampling sequence of no less than ten consecutive minutes.

[0030] The installation coordinates of each steel support are recorded synchronously to form an installation position information set, and the first stress time and installation completion time of each steel support are recorded as a construction time record set.

[0031] In this embodiment, strain gauge axial force sensors are arranged on each steel support, and each sensor is assigned a unique number. After low-pass anti-aliasing processing and analog-to-digital conversion, the sensors output temperature-compensated axial force values. Sampling employs an equal-slot strategy and timestamps are applied on a unified time baseline (preferably PTP or NTP, supplemented by local clock calibration) to ensure that the time alignment error between different sensors is within milliseconds. Sampling parameters are configurable, but each sensor should acquire at least 10 minutes of continuous equal-slot sequences. The acquisition results are saved as a structured time series—each record includes at least: sensor number, UTC timestamp, original strain or calibrated axial force value, temperature, status identifier, and checksum, for subsequent verification and traceability.

[0032] It is important to note that all timestamps should be stored in the database using the ISO-8601 (UTC) format. Any processing of the raw data (such as filtering, missing data rollback, imputation sampling, or window expansion) should create a new data version and record the processing method, processing time, executor, or algorithm version number in the metadata. Historical versions should be retained for subsequent comparison and tracing.

[0033] Simultaneously, the installation location information set is recorded based on engineering benchmarks. The coordinates are measured using RTK-GNSS or total station and stored in the record in the form of plane / engineering coordinates. At the same time, the location measurement accuracy estimate is saved. The construction time record set directly associates event semantics with the time series by inserting event identifiers (such as installation completion time, first stress time) into the time series and storing event types and times in the metadata.

[0034] Real-time integrity verification and fault-tolerant labeling are implemented in the transmission and storage process. Missing or abnormal samples are labeled and their original values ​​and repair records are recorded. For the subsequent extraction of baseline axial force reference values, candidate sample windows are selected from the axial force time series dataset according to the agreed window and the stability criterion is calculated (the stability criterion is defined as the ratio of the sample standard deviation to the sample mean). When the preset stability threshold is met, the typical statistics (such as the median or mean) of the sample window are recorded as the baseline axial force reference value and included in the baseline axial force reference value set. The axial force time series dataset, the installation location information set, and the construction time record set correspond to each other in terms of format and index to ensure subsequent point-by-point matching.

[0035] In one implementation of this invention, 30 steel supports are installed on-site, each equipped with a strain sensor (numbered 001 to 030). The sensors sample once per second (i.e., a sampling frequency of 1 Hz) for 10 minutes, generating 600 equally spaced data points per sensor. Hardware low-pass filtering is applied to the sensor terminals, and the data is quantized using a 24-bit analog-to-digital converter. Time synchronization uses PTP, ensuring that the timestamp error between different sensors is less than 1 millisecond. Installation location information is measured using RTK-GNSS and recorded as planar coordinates; for example, the measurement result for support 03 is (12.34 meters, 5.67 meters), with a measurement accuracy of ±0.05 meters. The installation completion time is written as "2025-09-01 10:05:00", and the first stress event is written as "2025-09-01 10:12:40". Baseline extraction example: The median of the data collected in the first 2 minutes after installation (120 points) is used as a candidate baseline. If the standard deviation of this window is 80 N and the sample mean is 12500 N, then the stability criterion is 80 ÷ 12500 = 0.0064, which is less than the assumed threshold of 0.05. Therefore, the median of 12500 N is accepted as the baseline axial force reference value for this support. A single-line example of the data file (CSV format, for ease of understanding) is as follows:

[0036] 003,2025-09-01T10:05:00.000Z,12500,22.3,OK,12.34,5.67,INSTALL_DONE,3A5F

[0037] The fields represent, in order: sensor number, UTC timestamp, axial force (N), temperature (degrees Celsius), status, coordinate X, coordinate Y, event identifier, and checksum.

[0038] The above example illustrates how to form three datasets from the perspectives of sensor, time synchronization, recording format, coordinate measurement, and event annotation: axial force time series dataset, installation location information set, and construction time record set.

[0039] Step S2: Calculate the baseline axial force reference value for each steel support based on the axial force time series dataset;

[0040] Preferably, step S2 includes:

[0041] Step S21: Extract the sample window from the axial force time series dataset before installation or during the initial idle period and calculate the median value;

[0042] Step S22: Using the median as the initial baseline estimate, calculate the stability coefficient within the window as the baseline stability criterion;

[0043] Step S23: When the baseline stability criterion is less than the preset threshold, the median value is determined as the baseline axial force reference value of the support; when it is greater than the preset threshold, the median value is recalculated using an extended sample window until the stability criterion is met, and the obtained median value is used as the baseline axial force reference value.

[0044] In this embodiment, the installation completion time or the first stress time marked in the construction time record set is used as the anchor point. An initial sample window is selected from the axial force time series dataset (the window length is determined by the configuration parameters, and two minutes can be used in the example). The sampling points in the sample window are subjected to integrity checks and missing or abnormally marked sample values ​​are removed. Then, the median value of the sample window is calculated as the robustness center quantity. At the same time, the arithmetic mean and sample standard deviation are calculated for stability determination.

[0045] Stability is determined by a stability coefficient (defined as the ratio of the sample standard deviation to the sample mean) compared with a preset threshold. When the stability coefficient is lower than the threshold, the median value of the sample window is recorded as the baseline axial force reference value for that support, and metadata such as the window start and end time, number of samples, missing rate, mean, standard deviation, and stability coefficient are written into the baseline record. When the stability coefficient is higher than the threshold, a window expansion strategy is triggered (the sample window is increased by a fixed step or by a multiple until the maximum configured length is reached). The integrity check and statistical calculation are repeated on the expanded window until the stability determination is met or the upper limit is reached. If the upper limit is reached but the stability is still not met, the support is marked as requiring manual review and the reason for failure is recorded. All intermediate statistics and the final baseline value are loaded into the baseline axial force reference value set, and a point-to-point traceable correspondence with the axial force time series dataset, installation location information set, and construction time record set is ensured by sensor identification and time index.

[0046] In one implementation of this invention, there are 30 supports on site, and the sensor samples once per second, with an initial window of 120 seconds (i.e., 120 sample points). The initial window statistical results for support 003 are: median 12500N, mean 12520N, standard deviation 80N, and stability coefficient 80÷12520=0.00639. If the stability threshold is set to 0.05, then 0.00639 is less than 0.05, so the median 12500N is accepted as the baseline for this support, and the start and end times of the window, sample size 120, missing rate 0%, mean 12520, standard deviation 80, and stability coefficient 0.00639 are recorded. If the initial window supporting 007 experiences significant fluctuations, with an initial window mean of 12500N, a standard deviation of 900N, and a stability coefficient of 900÷12500=0.07200 greater than 0.05, then the window is extended to 300 seconds (300 sample points) according to the strategy. After statistical analysis of the extended window, the median is 12530N, the mean is 12530N, the standard deviation is 100N, and the stability coefficient is 100÷12530=0.00798 less than 0.05. In this case, the median of 12530N is accepted as the baseline. Example of a single-line record (CSV, for easy display):

[0047] 003,2025-09-01T10:05:00Z,12500,12520,80,120,0.00639

[0048] The fields are, in order: sensor number, window start UTC time, median, mean, standard deviation, number of samples, and stability coefficient.

[0049] Step S3: Calculate the spatial attenuation coefficient of each steel support based on the installation location information set and the axial force time series dataset, and generate spatial attenuation coefficient data;

[0050] Preferably, step S3 includes:

[0051] Step S31: Calculate the Euclidean distance between adjacent supports of each steel support using the installation location information set and select the set of adjacent supports within the neighborhood radius;

[0052] Step S32: Calculate the spatial variation coefficient for the baseline axial force reference value and the current axial force difference of adjacent support sets;

[0053] Step S33: Calculate and normalize the axial force attenuation factor based on the difference between distance and axial force. The obtained axial force attenuation factor is used as the spatial attenuation coefficient of the support and recorded in the spatial attenuation coefficient data.

[0054] In this embodiment, based on the planar coordinates of each support in the installation location information set, the Euclidean distance between any two supports is calculated point by point, and a set of adjacent supports in the neighborhood is selected with a configurable neighborhood radius; for each adjacent support in the selected neighborhood, the baseline value is read from the baseline axial force reference value set, and the axial force value corresponding to the current observation time is read from the axial force time series data set, and the axial force difference of each adjacent support is calculated (the absolute value is taken to represent the deviation).

[0055] For each support, adjacent supports within a neighborhood are selected, and the coefficient of variation (standard deviation / mean) of the baseline values ​​within the neighborhood is calculated as a consistency criterion. For each neighboring point, its axial force deviation from the center point is converted into an "attenuation component" by applying a distance weight (e.g., negative exponential decay). The attenuation components of all neighboring points are summed and normalized by the sum of the absolute values ​​of the deviations within the neighborhood to obtain a scalarized spatial attenuation coefficient. In cases where the neighborhood is empty or the consistency criterion is abnormal, the system should mark the error and trigger a manual check.

[0056] The scalarization result is used as the spatial attenuation coefficient of the support and written into the spatial attenuation coefficient dataset. The dataset also records the neighborhood radius, distance attenuation scale, participating neighbor identifiers, neighbor distance sequence, baseline value sequence, observation value sequence, original attenuation components of each neighbor, aggregated attenuation amount and normalization result to ensure traceability point by point. When the neighborhood is empty, the sum of absolute differences is zero or the spatial variation coefficient exceeds the set range, the support is marked and manual review is triggered.

[0057] In one implementation of the present invention, the neighborhood radius of the evaluated support A is set to 3.0 meters, the distance attenuation scale λ is set to 1.5 meters, and there are three adjacent supports B, C, and D in the neighborhood, with measured Euclidean distances of 0.5 meters, 1.2 meters, and 2.5 meters, respectively; the corresponding baseline axial forces (N) are 12500, 12600, and 12400; the corresponding currently observed axial forces (N) are 12650, 12520, and 12350, respectively. Therefore, the absolute axial force differences of the three adjacent supports are 150, 80, and 50 N, respectively. First, the distance attenuation factor (negative exponential kernel) is calculated: exp(−0.5 / 1.5)=0.716531, exp(−1.2 / 1.5)=0.449329, exp(−2.5 / 1.5)=0.188876;

[0058] Multiplying the attenuation factor by their respective absolute differences yields three original attenuation components:

[0059] 150 × 0.716531 = 107.480, 80 × 0.449329 = 35.946, 50 × 0.188876 = 9.444;

[0060] Summing the three values ​​yields the aggregate attenuation: 107.480 + 35.946 + 9.444 = 152.870. Using the sum of the absolute differences within the neighborhood (150 + 80 + 50 = 280) as the normalization benchmark, we obtain the spatial attenuation coefficient: 152.870 ÷ 280 = 0.546 (approximately 0.546). This value is then entered into the spatial attenuation coefficient data.

[0061] In another way of explaining it: the above coefficient is approximately 0.546, which means that after considering distance attenuation, the effective influence of the axial force deviation observed in the neighborhood after distance attenuation accounts for about 54.6% of the total original deviation. This scalar is used for subsequent spatial modulation calculations or sorting and discrimination.

[0062] Preferably, the determination of the spatial variation coefficient in step S32 includes:

[0063] The ratio of the standard deviation to the mean of the baseline axial force reference value is calculated within a three-meter radius of the adjacent support set in the neighborhood as the spatial variation coefficient. When the spatial variation coefficient is between 0.20 and 0.70 and the baseline axial force shows a radial distribution with high center and low edge, the neighborhood influences the center and the supports in the neighborhood are included in the calculation range of the spatial attenuation coefficient.

[0064] In this embodiment, based on the installation location information set of the support to be judged, a neighborhood with a radius of three meters is selected, and the baseline axial force reference value sequence of all supports in the neighborhood is extracted. First, the integrity of the sequence is checked and missing or outlier values ​​are removed. Then, the arithmetic mean and sample standard deviation of the sequence are calculated, and the ratio of the sample standard deviation to the mean is used as the spatial variation coefficient. The obtained spatial variation coefficient is compared with the interval [0.20, 0.70]. If the coefficient is within the interval, the radial distribution characteristics are further calculated on the concentric ring with the support to be judged as the origin (for example, the neighborhood is divided into three rings: near distance, medium distance, and far distance: 0–0.8 meters, 0.8–1.8 meters, and 1.8–3.0 meters). The average value of the baseline axial force in each ring is calculated and the trend of the average value change between rings is compared.

[0065] When the radial average value shows a monotonically decreasing trend from the inside to the outside and the average value of the near ring is significantly higher than that of the far ring (e.g., the near average is at least 5% higher than the far average), it is determined to be a radial distribution pattern of "high center and low edge". Only when the spatial variation coefficient falls within the specified range and the radial distribution meets the condition of high center and low edge, it is confirmed that there is an influencing center in the neighborhood and each support in the neighborhood is included in the subsequent calculation range of spatial attenuation coefficient. At the same time, the participation point identifier, distance sequence, mean, standard deviation, spatial variation coefficient, average value of each ring and judgment result are written into the spatial variation coefficient data record to ensure traceability and quality control.

[0066] In one implementation of the present invention, there are four supports within the neighborhood of the support to be determined, with coordinate distances of S1 (0 meters, itself), S2 (0.6 meters), S3 (1.4 meters), and S4 (2.7 meters), respectively. The corresponding baseline axial force reference values ​​are 15000N, 12000N, 9000N, and 8000N. The average value of these four values ​​is calculated to be 11000N, the sample standard deviation is approximately 3162N, and the spatial variation coefficient is approximately 0.287 (3162 ÷ 11000 ≈ 0.287). The values ​​fall between 0.20 and 0.70. Divided by radial rings, the near-range ring (0–0.8 meters) includes 15000 and 12000, with a near-range average of 13500N; the mid-range ring has an average of 9000N; and the far-range ring has an average of 8000N. The average values ​​of the three rings decrease monotonically from the inside out, and the near-range average is about 68.7% higher than the far-range average, satisfying the condition of "high in the center and low at the edge". Therefore, S1 is confirmed as the neighborhood influence center, and S1, S2, S3, and S4 are included in the spatial attenuation coefficient calculation range.

[0067] Step S4: Calculate the time response coefficient of each steel support based on the construction time record set and the axial force time series dataset, and form a time response coefficient set;

[0068] Preferably, the calculation of the time response coefficient in step S4 includes:

[0069] Obtain the construction time record set and axial force time series dataset for each steel support, and calculate the number of days between the current observation time and the installation completion time;

[0070] When the interval is less than 3 days, the time response coefficient is set to a suppression coefficient in the range of 0.1 to 0.2.

[0071] When the interval is between 4 and 14 days, the time response coefficient is calculated based on the delayed Gaussian shape function, where the delay center is 10 days and the width is 5 days.

[0072] When the interval is between 15 and 30 days, the time response coefficient is set to the range of 0.8 to 0.95; the time response coefficient is recorded in the time response coefficient set.

[0073] In this embodiment, the installation completion time (ISO8601UTC format) of each support is read from the construction time record set and the difference is calculated with the current observation time to obtain the interval in days (the floating-point number of days is obtained by dividing the difference in seconds by 86400 and retaining three decimal places); missing or abnormal installation times are marked and rolled back or manual review is triggered according to the strategy.

[0074] The time response coefficients are determined segmented based on the number of days between the observation time and the installation completion time. For intervals less than 3 days, a lower suppression coefficient (approximately 0.10–0.20) is used; for intervals between 4–14 days, a continuous coefficient is generated using a delay morphology function centered at 10 days (e.g., a Gaussian template), then mapped to the target interval according to a pre-calibrated scale and pruned to upper and lower limits; for intervals between 15–30 days, a smooth linear mapping places the coefficients in the 0.80–0.95 range. All calculation results are accompanied by metadata (interval in days, method used, parameters, and quality indicators) for review. When the interval is between 15 and 30 days, the days are mapped to a fixed interval [0.80, 0.95] using a linear mapping (i.e., the coefficient smoothly increases from 0.80 to 0.95 as the number of days increases).

[0075] After each calculation, the time response coefficients are written to the time response coefficient set, along with metadata fields: sensor identifier, installation completion time, observation time, interval in days, method identifier, parameters used (delay center, width, upper and lower limits of mapping), original Gaussian value (if applicable), final coefficient value (retaining three decimal places), and quality flag (normal / rollback / requires review).

[0076] In one implementation of this invention, support A: installation completion time is 2025-09-01T10:00:00Z, current observation is 2025-09-02T22:00:00Z, interval is 1.5 days (1.500), which is less than 3 days. Using the linear interpolation coefficient = 0.10 + (1.5 / 3.0) × 0.10 = 0.150, an example of a record row is shown below:

[0077] A,2025-09-01T10:00:00Z,2025-09-02T22:00:00Z,1.500,LINEAR_<3d,params[lower=0.10,upper=0.20],coeff=0.150,flag=OK.

[0078] Support B: Installation completion time is 2025-08-25T09:00:00Z, current observation is 2025-09-02T09:00:00Z, interval is 8.0 days, falling within the 4–14 day range. First calculate the Gaussian value g≈exp(−(8−10)^2 / (2×5^2))≈exp(−0.08)≈0.923, and then use the mapping coefficient =0.20+(0.95−0.20)×g≈0.20+0.75×0.923≈0.892;

[0079] Support C: Installation completion time is 2025-08-10T12:00:00Z, current observation is 2025-09-01T12:00:00Z, the interval is 22.0 days, falling within the 15-30 day range, according to the linear mapping coefficient = 0.80+((22−15) / (30−15))×0.15=0.80+(7 / 15)×0.15=0.87.

[0080] It should be noted that the record row examples supporting B and supporting C are similar to those supporting A, and similar technical means will not be shown in this embodiment.

[0081] like Figure 2 The figure shown is a schematic diagram of the time response coefficient curve.

[0082] Preferably, in step S4, the calculation of the baseline axial force reference value, spatial attenuation coefficient data, time response coefficient set, and the currently observed axial force value according to a predetermined combination rule is specifically as follows:

[0083] Divide the baseline deviation of each support by 1000N to obtain the standardized first input term;

[0084] Use the spatial attenuation coefficient data directly as the second input item;

[0085] Use the product of the first input item and the second input item as the third input item;

[0086] The first input term is multiplied by 0.50 according to the pre-calibrated weights to obtain the structural contribution term, the second input term is multiplied by 0.30 to obtain the positional contribution term, and the third input term is multiplied by 0.20 to obtain the collaborative contribution term.

[0087] The structural contribution, location contribution, and collaborative contribution are added together to obtain the comprehensive response index. The comprehensive response index is multiplied by the spatial attenuation coefficient data and then by the time response coefficient to obtain the spatial-temporal modulated axial force increment. The axial force increment is added to the baseline axial force reference value to obtain the initial axial force estimate.

[0088] In this embodiment, the difference between the current observed axial force value and the baseline axial force reference value is used as the baseline deviation (signed difference). This deviation is divided by 1000N to obtain the standardized first input term. The spatial attenuation coefficient data is directly read as the second input term. The first input term and the second input term are multiplied to obtain the third input term. The first input term is multiplied by 0.50 according to the pre-calibrated weights to obtain the structural contribution term, the second input term is multiplied by 0.30 to obtain the position contribution term, and the third input term is multiplied by 0.20 to obtain the collaborative contribution term. The three terms are added together to obtain the dimensionless comprehensive response index. In order to convert the dimensionless response into an axial force increment with physical dimensions, the comprehensive response index is scaled according to the pre-calibrated conversion coefficient (system parameters, obtained through training or on-site calibration). Then, the scaled response is multiplied by the spatial attenuation coefficient and the time response coefficient in sequence to complete the space-time modulation and obtain the final axial force increment. The axial force increment is added to the baseline axial force reference value to obtain the initial value of the axial force estimation. Implementation details include: maintaining a consistent unit / dimension description for all intermediate quantities and recording metadata (sensor ID, timestamp, baseline deviation, first / second / third input item, each contribution item, comprehensive response index, weights used, conversion coefficient, spatial attenuation coefficient, time response coefficient, final increment, and quality indicator); implementing upper and lower limit pruning for outliers and writing processing tags; and standardizing numerical precision to three decimal places in the calculation chain for archiving and comparison.

[0089] In one implementation of the present invention, the baseline axial force reference value is 12500N; the currently observed axial force value is 12700N;

[0090] Therefore, the baseline deviation = 12700 − 12500 = 200N.

[0091] First input item = baseline deviation ÷ 1000 = 200 ÷ 1000 = 0.200.

[0092] The second input (spatial attenuation coefficient) = 0.546 (calculated from the neighborhood).

[0093] Third input item = First input item × Second input item = 0.200 × 0.546 = 0.109200.

[0094] Structural contribution = First input item × 0.50 = 0.200 × 0.50 = 0.100.

[0095] Location contribution = Second input item × 0.30 = 0.546 × 0.30 = 0.163800.

[0096] Collaborative contribution item = third input item × 0.20 = 0.109200 × 0.20 = 0.021840.

[0097] The overall response index = 0.100 + 0.163800 + 0.021840 = 0.285640.

[0098] Set the pre-calibration conversion factor (scale factor) to 1000 N Newtons per dimensionless unit (this value is determined by the calibration process and written into the system parameters), and the time response coefficient to 0.892 (from the time response set).

[0099] Then the axial force increment = comprehensive response index × spatial attenuation coefficient × time response coefficient × conversion coefficient = 0.285640 × 0.546 × 0.892 × 1000 ≈ 139.116 N (rounded to three decimal places).

[0100] Initial value of axial force estimation = baseline axial force reference value + axial force increment = 12500 + 139.116 = 12639.116 N.

[0101] Step S5: Calculate the baseline axial force reference value, spatial attenuation coefficient data, time response coefficient set and the current observed axial force value according to the predetermined combination rules to obtain the initial axial force estimate value of each steel support and generate the axial force estimate value.

[0102] Preferably, the method of adding the axial force increment to the baseline axial force reference value to obtain the initial axial force estimate further includes:

[0103] The initial value of the axial force estimate, the difference with the baseline axial force reference value, the spatial attenuation coefficient value, and the time response coefficient value corresponding to each steel support are extracted one by one and arranged in the same data record to obtain the corresponding four joint feature data.

[0104] Representative support points were selected within the prefabricated station to collect measured axial forces to form a calibration dataset. The calibration dataset was then paired with the corresponding four joint feature data to form training sample pairs.

[0105] The axial force inversion relationship is established using training sample pairs, and the measured axial force is used as the target value. The four joint feature data are used as input variables for the axial force inversion relationship, and the inverted axial force value of each support is output.

[0106] In this embodiment, the initial axial force estimation records are read one by one and aligned with the sensor identifier and timestamp. Fields are extracted item by item (initial axial force estimation value, baseline axial force reference value and its difference, spatial attenuation coefficient, and time response coefficient). Each field undergoes integrity and consistency checks (field existence, unit in Newtons (N), numerical precision to three decimal places, missing data annotation, or backtracking based on the most recent valid value and recording the reason for backtracking). The four features are arranged sequentially in the same data row in a fixed order and metadata (sensor ID, UTC timestamp, coordinates, and data quality flag) is appended. In the calibration dataset formed by field sampling, each measured axial force record is matched using a time proximity rule (selected from the sampling time). The data with the smallest absolute difference (not exceeding a preset threshold, such as a five-minute time window) is paired with the corresponding four joint feature data to form a training sample pair table. The training sample pair table is stored in a structured table format (such as Parquet or CSV with schema), and the features are uniformly transformed and normalized before being entered into the table (using the training set mean and standard deviation for standardization or using predefined dimensional scaling). At the same time, the normalized statistics used are saved as metadata for reuse during inference. To establish the axial force inversion relationship, a linear combination framework is adopted and the weight coefficients and constant terms are solved using the method of minimizing the sum of squared errors. During the solution process, the loss curve, fitting coefficients and version number are recorded, and the obtained parameters are written into the parameter table.

[0107] In one implementation of the present invention, the record of support number 003 at a certain moment is as follows: the baseline axial force reference value is 12500.000N, and the initial axial force estimation value is 12639.116N; first, the baseline deviation is calculated: 12639.116 − 12500.000 = 139.116N; this deviation and the other two items are extracted one by one into four joint feature data (the field order is fixed as: initial axial force estimation value, baseline deviation, spatial attenuation coefficient, and time response coefficient), assuming that the spatial attenuation coefficient is 0.546 and the time response coefficient is 0.892.

[0108] During on-site calibration, the measured axial force of the support was 12640.000N at 2025-09-01T10:12:45Z (5 seconds difference from the recorded time, ≤300 seconds for pairing).

[0109] If a training table is formed from several such samples (e.g., 100 records), the weight coefficients of the linear mapping can be estimated by minimizing the sum of squared errors and the results can be written into the parameter table. The parameter table, together with the normalized statistics, is used to perform real-time inversion on new records and ensure that the results are traceable.

[0110] Most importantly, in the process of forming training sample pairs, it is essential to ensure that the baseline axial force reference value, spatial attenuation coefficient value, time response coefficient value, and initial axial force estimation value in the four joint feature data have been generated and matched one by one with the measured axial force in the training sample pairs.

[0111] It is important to note that the four joint feature data used to constitute the training sample pairs (initial value of axial force estimation, difference of baseline axial force reference value, spatial attenuation coefficient, and time response coefficient) must be directly derived from the data records generated and archived in steps S2, S3, S4, and S5. Furthermore, the four features in each training sample pair should correspond to the corresponding measured axial force in terms of timestamp, sensor identifier, and coordinates. If any feature is a rollback or repair value, it must be clearly marked in the sample pair metadata (including the reason for rollback and the original value) to ensure the traceability and auditability of the training samples.

[0112] Preferably, the four joint feature data used as input variables for the axial force inversion relationship include:

[0113] Using training sample pairs as input, the weight coefficients and constant terms in the axial force inversion relation are solved by minimization method, and the axial force inversion relation is expressed in linear combination form to obtain the final inversion weight coefficients.

[0114] The axial force inversion relationship is applied to the four joint characteristic data of all steel supports in the prefabricated station, and the inversion axial force value of each support is output.

[0115] In this embodiment, the training samples are represented as a feature matrix and a target vector. A design matrix X is constructed according to a fixed field order (each row contains a constant term and four joint features). The target vector y is the corresponding measured axial force value. Before data storage, consistency checks are performed on X and y (time pairing threshold, unit consistency, missing and anomaly handling strategies). A defined transformation or standardization is applied to numerical scale differences, and the transformation parameters are recorded as metadata. The ordinary least squares method is used to minimize the sum of squared residuals to solve for the weight vector (analytical solutions or numerically stable QR / SVD solvers can be used, and the matrix condition number is recorded during the solution to verify numerical stability). The analytical solution is in the form of a coefficient vector obtained by solving the normal equation (recorded in the model parameter table along with the number of training samples, training / validation segmentation information, training loss, residual standard deviation, and fitting diagnostic indicators such as root mean square error and coefficient of determination).

[0116] To prevent overfitting, least squares with regularization terms can be enabled as needed, and the regularization parameters can be recorded. After training, the model parameters (including constant terms and feature weights), the mean and standard deviation used for normalization, the solution method and version number are written into the parameter storage, and the residual distribution is calculated on the independent validation set to estimate the prediction uncertainty. When the model is applied to the entire site (inference stage), the four joint feature data of each support are read one by one, the input is transformed according to the same missing data handling and normalization steps as during training, the predicted value is calculated (i.e., the constant term plus the product of each weight and the corresponding feature), the predicted output is denormalized to restore the physical units, the prediction residual and the confidence interval based on the standard deviation of the training residual are calculated and written into the output table, and a quality flag is marked for records that exceed the confidence interval or contain abnormal input.

[0117] In one implementation of this invention, the model parameters (example values) have been trained as follows: constant term 211.299, weight pair initial estimate 0.97, weight pair baseline deviation 0.50, weight pair spatial decay coefficient 100.0, and weight pair time response coefficient 50.0. For the four joint features supporting 003 at time t: initial estimate 12639.116N, baseline deviation 139.116N, spatial decay coefficient 0.546, and time response coefficient 0.892, these are directly substituted and calculated item by item using the same unnormalized training path.

[0118] The initial term multiplied by the weight is 12639.116 × 0.97 = 12259.94252 (approximately 12259.943).

[0119] Baseline deviation term: 139.116 × 0.50 = 69.558;

[0120] Spatial term: 0.546 × 100.0 = 54.600; Time term: 0.892 × 50.0 = 44.600;

[0121] Adding all the above items together and adding the constant term 211.299: 12259.94252 + 69.558 = 12329.50052;

[0122] 12329.50052 + 54.600 = 12384.10052;

[0123] 12384.10052 + 44.600 = 12428.70052;

[0124] 12428.70052 + 211.299 = 12639.99952 (approximately 12640.000), yielding an inverted axial force value of approximately 12640.000 N.

[0125] Preferably, the output of the inverted axial force values ​​for each support also includes:

[0126] The axial force anomaly index is obtained by calculating the point-by-point difference between the inverted axial force values ​​of all supports in the prefabricated station and the baseline axial force reference values.

[0127] Support data for axial force anomaly indices exceeding preset anomaly index thresholds are used to generate an anomaly alarm list;

[0128] Based on the abnormal alarm list, a priority handling list is output by sorting the spatial attenuation coefficient data, and a visualization chart of steel support axial force sensing data analysis is generated.

[0129] In this embodiment, the inverted axial force value and the corresponding baseline axial force reference value of each support in the result table are read one by one, and the point-by-point axial force anomaly index is calculated (represented by relative difference, that is, the dimensionless ratio is obtained by subtracting the baseline axial force reference value from the inverted axial force value and then dividing by the baseline axial force reference value). Before the calculation, data quality verification is performed (checking timestamp matching, data integrity flags and numerical ranges, marking missing or outlier values ​​and backing up or skipping them according to rules); short-term smoothing or denoising (e.g., moving median filtering, with the window length configured according to the sampling frequency) is applied to the anomaly index sequence that comes in in real time to reduce the instantaneous noise trigger rate.

[0130] When the smoothed anomaly index exceeds the preset anomaly index threshold and meets the persistence condition (e.g., the number of observation points continuously exceeding the threshold or the duration reaching the configured minimum window), the support is recorded as an anomaly and a record is generated in the anomaly alarm list. The alarm record includes the following fields: support identifier, alarm time, inverted axial force value, baseline axial force reference value, anomaly index, spatial attenuation coefficient value, time response coefficient, continuous over-threshold count, and data quality flag. The generated anomaly alarm list is sorted in descending order of spatial attenuation coefficient value (those with larger spatial attenuation coefficients are included in the priority handling list first). In the case of ties, the size of the anomaly index is used as the secondary sorting basis, and the sorting reason is recorded. All alarms and sorting results are written into a structured table and a visualization output is generated simultaneously (including: single-point time series diagram - time on the horizontal axis, axial force and baseline on the vertical axis, and anomaly time marked; plan layout diagram - the anomaly index is represented by color gradient or symbol size according to the installation location information; priority handling bar chart - the supports and their spatial attenuation coefficients and anomaly indices are listed in order; alarm list table).

[0131] In one implementation of the present invention, the preset anomaly index threshold is set to 0.02 (i.e., 2%), and the persistence requirement is at least three consecutive time slots exceeding the threshold; the supporting baseline A is 12500N, the inversion value is 12639.116N, and the anomaly index = (12639.116−12500)÷12500≈0.0111 (1.11%, lower than 0.02, no alarm); the supporting baseline B is 12500N, the inversion value is 12875N, and the anomaly index = (12875−12500)÷12500=0.03 (3%, >0.02).

[0132] If three consecutive observations exceed the threshold, a record will be generated in the abnormal alarm list.

[0133] B,ts,12875,12500,0.030,spatial_coef=0.800,time_coeff=0.870,persist=3,quality=OK;

[0134] The baseline for support C is 12500N, the inversion value is 12700N, and the anomaly index is 0.016 (1.6%, not exceeding the threshold). After sorting the alarm list in descending order of spatial attenuation coefficient, if only B is abnormal, then the priority handling list is [B]. At the same time, the following are generated: a time series diagram of B (showing the inversion value and baseline, and marking the three threshold exceeding points in red), and point B is marked with a dark color on the plan view.

[0135] It is important to note that when sorting the abnormal alarm list by spatial attenuation coefficient and outputting the priority handling list, the system should simultaneously generate a sorting mapping table. This mapping table lists the source fields used for sorting for each alarm record (at least including: baseline axial force reference value, inverted axial force value, the difference between the two, spatial attenuation coefficient, time response coefficient, original abnormal sequence used to determine persistence and its time window), and writes the mapping table and the alarm list into the archive with the same index.

[0136] Most importantly, after generating the axial force anomaly index and forming an anomaly alarm list, the spatial attenuation coefficient data must be called to perform a secondary sorting of the anomaly alarm list. During the sorting process, the difference between the corresponding baseline axial force reference value and the inverted axial force value is recorded item by item to form a mapping table between the sorting results and the original data.

[0137] It is particularly noteworthy that all numerical items in the text (axial force, deviation, coefficient, etc.) must have consistent units and precision standards in different processing stages (axial force is in Newtons (N) and retained to three decimal places; coefficients are dimensionless and retained to three decimal places). The system must specify the unit and decimal precision of the fields in each intermediate table and the final table. The calculation results should first be checked for reasonableness (e.g., upper limit of baseline deviation, range of spatial attenuation coefficient, range of time response coefficient). Records that exceed the reasonable range must be marked and enter the manual review process.

[0138] This embodiment is only used to illustrate the implementation of the present invention and does not limit the scope of protection of the invention. Any equivalent substitutions or improvements made to the implementation methods within the principles and spirit of the present invention should be included within the scope of protection of the present invention.

[0139] like Figure 3 As shown, by combining the results of S2, S3, and S5, a 5×5 grid is constructed, with each grid point representing a steel support, to collectively represent a grayscale simulation diagram of the axial force value distribution of the station supports.

[0140] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for intelligent analysis of axial force sensing data of steel supports in prefabricated railway stations, characterized in that, Includes the following steps: Step S1: Collect the axial force time series dataset, installation location information set, and construction time record set for each steel support in the prefabricated station; Step S2: Calculate the baseline axial force reference value for each steel support based on the axial force time series dataset; Step S3: Calculate the spatial attenuation coefficient of each steel support based on the installation location information set and the axial force time series dataset, and generate spatial attenuation coefficient data; Step S3 includes: Step S31: Calculate the Euclidean distance between adjacent supports of each steel support using the installation location information set and select the set of adjacent supports within the neighborhood radius; Step S32: Calculate the spatial variation coefficient for the baseline axial force reference value and the current axial force difference of adjacent support sets; Step S33: Calculate and normalize the axial force attenuation factor based on the difference between distance and axial force. The obtained axial force attenuation factor is used as the spatial attenuation coefficient of the support and recorded in the spatial attenuation coefficient data. Step S4: Calculate the time response coefficient of each steel support based on the construction time record set and the axial force time series dataset, and form a time response coefficient set; Step S4 includes: Obtain the construction time record set and axial force time series dataset for each steel support, and calculate the number of days between the current observation time and the installation completion time; When the interval is less than 3 days, the time response coefficient is set to a suppression coefficient in the range of 0.1 to 0.

2. When the interval is between 4 and 14 days, the time response coefficient is calculated based on the delayed Gaussian shape function, where the delay center is 10 days and the width is 5 days. When the interval is between 15 and 30 days, set the time response coefficient to the range of 0.8 to 0.95; record the time response coefficient in the time response coefficient set; Step S5: Calculate the baseline axial force reference value, spatial attenuation coefficient data, time response coefficient set and the current observed axial force value according to the predetermined combination rules to obtain the initial axial force estimate value of each steel support and generate the axial force estimate value.

2. The intelligent analysis method for axial force sensing data of steel supports for prefabricated railway stations according to claim 1, characterized in that, Step S1 includes: The distributed strain gauge axial force sensor network is controlled to collect axial force time series data of each steel support under static load and working load according to a unified time baseline. The axial force time series dataset contains an equal time-slot sampling sequence of no less than ten consecutive minutes. The installation coordinates of each steel support are recorded synchronously to form an installation position information set, and the first stress time and installation completion time of each steel support are recorded as a construction time record set.

3. The intelligent analysis method for axial force sensing data of steel supports for prefabricated railway stations according to claim 1, characterized in that, Step S2 includes: Step S21: Extract the sample window from the axial force time series dataset before installation or during the initial idle period and calculate the median value; Step S22: Using the median as the initial baseline estimate, calculate the stability coefficient within the window as the baseline stability criterion; Step S23: When the baseline stability criterion is less than the preset threshold, the median value is determined as the baseline axial force reference value of the support; when it is greater than the preset threshold, the median value is recalculated using an extended sample window until the stability criterion is met, and the obtained median value is used as the baseline axial force reference value.

4. The intelligent analysis method for axial force sensing data of steel supports for prefabricated railway stations according to claim 1, characterized in that, The determination of the spatial variation coefficient in step S32 includes: The ratio of the standard deviation to the mean of the baseline axial force reference value is calculated within a three-meter radius of the adjacent support set in the neighborhood as the spatial variation coefficient. When the spatial variation coefficient is between 0.20 and 0.70 and the baseline axial force shows a radial distribution with high center and low edge, the neighborhood influences the center and the supports in the neighborhood are included in the calculation range of the spatial attenuation coefficient.

5. The intelligent analysis method for axial force sensing data of steel supports for prefabricated railway stations according to claim 1, characterized in that, Step S4 involves calculating the baseline axial force reference value, spatial attenuation coefficient data, time response coefficient set, and the currently observed axial force value according to a predetermined combination rule. Divide the baseline deviation of each support by 1000N to obtain the standardized first input term; Use the spatial attenuation coefficient data directly as the second input item; Use the product of the first input item and the second input item as the third input item; The first input term is multiplied by 0.50 according to the pre-calibrated weights to obtain the structural contribution term, the second input term is multiplied by 0.30 to obtain the positional contribution term, and the third input term is multiplied by 0.20 to obtain the collaborative contribution term. The structural contribution, location contribution, and collaborative contribution are added together to obtain the comprehensive response index. The comprehensive response index is multiplied by the spatial attenuation coefficient data and then by the time response coefficient to obtain the spatial-temporal modulated axial force increment. The axial force increment is added to the baseline axial force reference value to obtain the initial axial force estimate.

6. The intelligent analysis method for axial force sensing data of steel supports for prefabricated railway stations according to claim 5, characterized in that, The initial axial force estimation value is obtained by adding the axial force increment to the baseline axial force reference value, and also includes: The initial value of the axial force estimate, the difference with the baseline axial force reference value, the spatial attenuation coefficient value, and the time response coefficient value corresponding to each steel support are extracted one by one and arranged in the same data record to obtain the corresponding four joint feature data. Representative support points were selected within the prefabricated station to collect measured axial forces to form a calibration dataset. The calibration dataset was then paired with the corresponding four joint feature data to form training sample pairs. The axial force inversion relationship is established using training sample pairs, and the measured axial force is used as the target value. The four joint feature data are used as input variables for the axial force inversion relationship, and the inverted axial force value of each support is output.

7. The intelligent analysis method for axial force sensing data of steel supports for prefabricated railway stations according to claim 6, characterized in that, The four joint characteristic data points used as input variables for the axial force inversion relationship include: Using training sample pairs as input, the weight coefficients and constant terms in the axial force inversion relation are solved by minimization method, and the axial force inversion relation is expressed in linear combination form to obtain the final inversion weight coefficients. The axial force inversion relationship is applied to the four joint characteristic data of all steel supports in the prefabricated station, and the inversion axial force value of each support is output.

8. The intelligent analysis method for axial force sensing data of steel supports for prefabricated railway stations according to claim 7, characterized in that, The output of the inverted axial force values ​​for each support also includes: The axial force anomaly index is obtained by calculating the point-by-point difference between the inverted axial force values ​​of all supports in the prefabricated station and the baseline axial force reference values. Support data for axial force anomaly indices exceeding preset anomaly index thresholds are used to generate an anomaly alarm list; Based on the abnormal alarm list, a priority handling list is output by sorting the spatial attenuation coefficient data, and a visualization chart of steel support axial force sensing data analysis is generated.

Images