Intelligent identification and early warning system and method for abnormal working conditions of pile foundation construction

Abstract

This invention discloses an intelligent identification and early warning system and method for abnormal conditions in pile foundation construction, relating to the field of civil engineering technology. The method includes: acquiring standard historical multi-source data, generating time-frequency component data through date classification, FFT, and wavelet transform; setting differentiated rolling windows based on the time-frequency components, and obtaining a dynamic baseline through weighted fusion; filtering effective data to calculate the coefficient of variation, determining the optimal sensitivity parameter to generate a dynamic early warning threshold; completing short-term construction parameter prediction by collecting real-time data, constructing a topology structure and LSTM model; calculating FCR and LSC values, and dividing the threshold exceedance range into four levels of early warning and generating strategies. This technology integrates multi-source data and time-frequency analysis with deep learning prediction technology, improving the accuracy, real-time performance, and adaptability of early warnings, reducing the risk of false alarms and missed alarms, and providing support for refined safety management of pile foundation construction.

Description

Technical Field

[0001] This invention belongs to the field of civil engineering technology, specifically relating to an intelligent identification and early warning system and method for abnormal working conditions in pile foundation construction. Background Technology

[0002] Pile foundation construction is one of the core processes in civil engineering. Its construction quality and safety directly determine the stability of the overall engineering structure. During construction, it is easily affected by multiple factors such as geological conditions, construction environment, and equipment status. The risk of various abnormal working conditions is high, and if not handled in a timely manner, it can easily lead to safety accidents, causing casualties and economic losses. Therefore, accurate monitoring and intelligent early warning of the pile foundation construction process are crucial. Traditional pile foundation construction monitoring relies heavily on manual inspections and single sensor data collection, which makes it difficult to achieve collaborative analysis of multi-source information and has poor adaptability to complex working conditions.

[0003] With the development of intelligent construction technology, the industry has gradually introduced data-driven monitoring and early warning methods, but existing technologies still have significant limitations. Some solutions only determine thresholds for single construction parameters, lacking in-depth analysis of the time-frequency and spatial correlation characteristics of these parameters, making it difficult to capture potential precursors to abnormal working conditions. At the same time, the dynamic baseline and early warning threshold settings lack flexibility and cannot adaptively adjust according to changes in the construction environment and working conditions, resulting in insufficient accuracy of early warnings and prominent issues of false alarms and missed alarms.

[0004] Currently, the integration of civil engineering with artificial intelligence and data processing technologies is becoming a trend, creating an urgent need for an early warning solution that integrates multi-source data, deep analysis techniques, and dynamic adaptation strategies to overcome the bottlenecks of traditional monitoring models. By integrating multi-dimensional construction data and combining time-frequency analysis, deep learning, and other technologies to achieve accurate prediction of working conditions and anomaly identification, the real-time performance, adaptability, and reliability of the early warning system can be improved. This provides technical support for the refined management and control of pile foundation construction safety, and has become an urgent need for industry development. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the existing technology, the first objective of this invention is to provide an intelligent identification and early warning system for abnormal working conditions in pile foundation construction; the second objective of this invention is to provide an intelligent identification and early warning method for abnormal working conditions in pile foundation construction.

[0006] The first objective of this invention is achieved through the following technical solution:

[0007] An intelligent identification and early warning system for abnormal conditions in pile foundation construction includes:

[0008] Historical data processing module: used to acquire standard historical multi-source pile foundation construction data, classify it by date, and generate time-frequency component data of pile foundation construction parameters through FFT and wavelet transform;

[0009] Dynamic baseline generation module: used to set a differentiated rolling window based on time-frequency component data, and obtain dynamic baseline value data for pile foundation construction by weighted fusion;

[0010] Threshold calculation module: used to filter valid historical data and calculate the coefficient of variation, determine the optimal alarm sensitivity parameters, and generate dynamic construction early warning threshold curve data;

[0011] The working condition prediction module is used to collect real-time data and construct the topology of pile foundation monitoring resources. It performs short-term construction parameter prediction based on the LSTM model and outputs short-term construction parameter prediction data.

[0012] Early warning triggering module: used to calculate the characteristic co-abnormality coefficient FCR value and the pile stability characteristic coefficient LSC value, and combine the dynamic early warning threshold and the over-limit amplitude to divide the early warning level and generate a multi-level early warning triggering strategy for pile foundation.

[0013] The second objective of this invention is achieved through the following technical solution:

[0014] A method for intelligent identification and early warning of abnormal working conditions in pile foundation construction is provided to realize an intelligent identification and early warning system for abnormal working conditions in pile foundation construction. The method includes the following steps:

[0015] S1: Obtain standard historical multi-source pile foundation construction data; classify the standard historical multi-source pile foundation construction data according to the historical construction monitoring date, perform time-frequency component decomposition of construction parameters, and generate time-frequency component data of pile foundation construction parameters;

[0016] S2: Based on the time-frequency component data of pile foundation construction parameters, a rolling window for construction parameters is set, and dynamic baseline values ​​are weighted and fused to obtain dynamic baseline value data for pile foundation construction.

[0017] S3: Perform effective data screening and coefficient of variation calculation on standard historical multi-source pile foundation construction data to generate effective historical pile foundation construction coefficient of variation; conduct sensitivity parameter analysis of early warning sources based on effective historical pile foundation construction coefficient of variation to determine the optimal alarm sensitivity parameter; combine effective historical pile foundation construction coefficient of variation and optimal alarm sensitivity parameter to perform adaptive abnormal early warning threshold processing on dynamic baseline value data of pile foundation construction to obtain dynamic construction early warning threshold curve data.

[0018] S4: Collect real-time status data of pile foundation construction equipment; construct a pile foundation monitoring resource topology based on the real-time pile foundation construction equipment status data, and generate pile foundation monitoring resource topology data; construct a pile foundation construction condition prediction model by combining the pile foundation monitoring resource topology data and historical multi-source time series data; use the model to predict short-term construction condition parameters and obtain short-term construction parameter prediction data.

[0019] S5: Calculate the FCR value based on short-term construction parameter prediction data and real-time image features, and combine the dynamic construction early warning threshold curve data to calculate the threshold condition difference and determine the multi-level early warning trigger, forming a multi-level early warning trigger strategy for pile foundations.

[0020] Preferably, step S1 includes the following sub-steps:

[0021] S11: Collect historical multi-source pile foundation construction data. The historical multi-source pile foundation construction data comes from various construction monitoring related databases and sensor historical records. The data types include construction parameters and corresponding image features for different geological types and different pile types.

[0022] S12: Perform multi-source format unification and preprocessing on historical multi-source pile foundation construction data to generate standard historical multi-source pile foundation construction data;

[0023] S13: Classify the standard historical multi-source pile foundation construction data according to the historical construction monitoring date to obtain classified historical multi-source pile foundation construction data;

[0024] S14: Perform frequency domain transformation on the classified historical multi-source pile foundation construction data to generate historical pile foundation construction spectrum data;

[0025] S15: Based on historical pile foundation construction spectrum data, perform time-frequency component decomposition of construction parameters to generate time-frequency component data of pile foundation construction parameters; the time-frequency component decomposition adopts wavelet transform, and the main fluctuation period of construction parameters is determined based on spectrum analysis during the decomposition process, and the appropriate wavelet basis function and decomposition level are selected accordingly.

[0026] Preferably, step S2 includes the following sub-steps:

[0027] S21: Based on the time-frequency component data of pile foundation construction parameters, set the construction parameter rolling window to obtain the corresponding rolling window for each time-frequency component;

[0028] S22: Calculate the baseline data of the corresponding time-frequency components through each rolling window;

[0029] S23: Based on the influence weight of each component on construction safety, assign dynamic weights to each baseline component and perform weighted fusion to obtain dynamic baseline value data for pile foundation construction; the weights are determined through expert experience or statistical analysis of historical data.

[0030] Preferably, step S3 includes the following sub-steps:

[0031] S31: Obtain the real-time pile foundation monitoring timestamp;

[0032] S32: Based on the real-time pile foundation monitoring timestamp, filter the valid data in the standard historical multi-source pile foundation construction data, calculate the coefficient of variation, and generate the valid historical construction coefficient of variation of pile foundations;

[0033] S33: Based on the effective historical construction variation coefficient of pile foundation, conduct sensitivity parameter analysis of early warning source to determine the optimal alarm sensitivity parameter; during the analysis process, set the initial sensitivity parameter in combination with the variation coefficient range, calculate the alarm performance index under different sensitivity parameters through backtracking test, and select the parameter with the best performance as the optimal alarm sensitivity parameter;

[0034] S34: Calculate the construction early warning threshold by combining the effective historical construction variation coefficient of the pile foundation with the optimal alarm sensitivity parameter, set the threshold constraint range based on engineering specifications and safety thresholds, and apply adaptive threshold constraints to the dynamic baseline value data of pile foundation construction to obtain dynamic construction early warning threshold curve data.

[0035] Preferably, step S32 includes the following sub-steps:

[0036] S321: Determine the effective historical construction window based on the real-time pile foundation monitoring timestamp to obtain effective historical construction window data;

[0037] S322: Filter standard historical multi-source pile foundation construction data through effective historical construction window data, and obtain effective historical pile foundation construction data after removing outliers;

[0038] S323: Calculate the mean and standard deviation based on effective historical construction data of pile foundations;

[0039] S324: Calculate the coefficient of variation by combining the mean and standard deviation to generate the effective historical construction coefficient of variation for pile foundations.

[0040] Preferably, in step S33, statistical analysis is performed on the effective historical construction variation coefficient of the pile foundation to observe the distribution characteristics, and parameter sub-ranges are divided according to the variation coefficient interval. Narrower parameter sub-ranges are set for sensitive parameters. During backtesting, real abnormal alarm and false alarm records are extracted, the initial sensitivity parameter range is traversed and the corresponding alarm performance index is calculated, and the parameter with the best early warning performance of the sensitive parameter is selected first.

[0041] Preferably, step S4 includes the following sub-steps:

[0042] S41: Collect real-time status data of pile foundation construction equipment and synchronous image features; image features are extracted using image processing technology;

[0043] S42: Construct a pile foundation monitoring resource topology based on real-time pile foundation construction equipment status data, use a graph data structure to describe the connection relationship and data transmission path of each component, and generate pile foundation monitoring resource topology data;

[0044] S43: Based on the topology of pile foundation monitoring resources, determine the time series parameters corresponding to the core monitoring nodes, and combine synchronous image features and topological spatial features to construct historical multi-source time series input samples using the sliding window method;

[0045] S44: An LSTM model is constructed based on multi-source temporal features as a prediction model for pile foundation construction conditions to be trained. The model contains an input layer, multiple LSTM hidden layers and a fully connected output layer. Regularization is used to suppress overfitting. The model is compiled through an adapter optimizer, and the loss function is selected to be suitable for continuous value prediction.

[0046] S45: Perform time-series sample processing on standard historical multi-source pile foundation construction data, divide the training set and test set and normalize them, iteratively train the LSTM model to the preset accuracy, and obtain the pile foundation construction condition prediction model.

[0047] S46: Use the pile foundation construction condition prediction model to predict short-term construction parameters and obtain short-term construction parameter prediction data.

[0048] Preferably, in step S5, the feature co-anomaly coefficient FCR and the pile stability feature coefficient LSC are calculated. A multi-level early warning system is constructed by combining the parameter over-limit amplitude, FCR value, image features and LSC value, and each level corresponds to a differentiated execution strategy.

[0049] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0050] 1. This invention integrates multi-source data such as drill pipe torque, mud properties, borehole wall displacement, and image features, and combines time-frequency component decomposition using FFT / wavelet transform with differentiated rolling window dynamic baseline generation to capture periodic fluctuations and sudden anomalies in construction parameters, reducing the risk of misjudgment from a single data source and improving early warning accuracy. Based on the LSTM model constructed from the topology of pile foundation monitoring resources, it predicts future trends in construction parameters in real time. Combined with a dynamic construction early warning threshold curve and a multi-level early warning triggering strategy using FCR / LSC values, it achieves graded responses from minor anomalies to extreme risks, improving the real-time performance of early warnings and the efficiency of construction safety management.

[0051] 2. This invention calculates the coefficient of variation by screening valid historical data and dynamically adjusts the early warning threshold by combining the optimal alarm sensitivity parameter. It adapts to different geological conditions, construction stages and environmental changes, balances recall and precision, reduces the risk of false alarms and missed alarms caused by parameter fluctuations or image interference, and supports refined decision-making for construction safety. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 The following is a block diagram of the intelligent identification and early warning system for abnormal conditions in pile foundation construction according to the present invention;

[0054] Figure 2 A flowchart of the intelligent identification and early warning method for abnormal working conditions in pile foundation construction according to the present invention is shown;

[0055] Figure 3 A flowchart of S4 of the present invention is shown. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of exemplary embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, steps, etc., can be employed. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0058] Example 1:

[0059] See Figure 1 As shown, the intelligent identification and early warning system for abnormal conditions in pile foundation construction in this embodiment includes:

[0060] Historical data processing module: used to acquire standard historical multi-source pile foundation construction data, classify it by date, and generate time-frequency component data of pile foundation construction parameters through FFT and wavelet transform;

[0061] Dynamic baseline generation module: used to set a differentiated rolling window based on time-frequency component data, and obtain dynamic baseline value data for pile foundation construction by weighted fusion;

[0062] Threshold calculation module: used to filter valid historical data and calculate the coefficient of variation, determine the optimal alarm sensitivity parameters, and generate dynamic construction early warning threshold curve data;

[0063] The working condition prediction module is used to collect real-time data and construct the topology of pile foundation monitoring resources. It performs short-term construction parameter prediction based on the LSTM model and outputs short-term construction parameter prediction data.

[0064] Early warning triggering module: used to calculate FCR and LSC values, combine dynamic early warning thresholds and over-limit amplitudes to classify early warning levels, and generate a multi-level early warning triggering strategy for pile foundations.

[0065] The beneficial effects of this embodiment are: intelligent identification and early warning of abnormal working conditions in pile foundation construction are achieved through multi-module collaboration; by combining FFT / wavelet transform, dynamic baseline, LSTM prediction and multi-level threshold triggering strategy, the accuracy, real-time performance and adaptability of early warning are improved, and the risk of false alarms and missed alarms are reduced.

[0066] Example 2:

[0067] See Figure 2 As shown in this embodiment, the intelligent identification and early warning method for abnormal working conditions in pile foundation construction is as follows:

[0068] S1: Obtain standard historical multi-source pile foundation construction data; classify the standard historical multi-source pile foundation construction data according to the historical construction monitoring date, perform time-frequency component decomposition of construction parameters, and generate time-frequency component data of pile foundation construction parameters;

[0069] S2: Based on the time-frequency component data of pile foundation construction parameters, a rolling window for construction parameters is set, and dynamic baseline values ​​are weighted and fused to obtain dynamic baseline value data for pile foundation construction.

[0070] S3: Perform effective data screening and coefficient of variation calculation on standard historical multi-source pile foundation construction data to generate effective historical pile foundation construction coefficient of variation; conduct sensitivity parameter analysis of early warning sources based on effective historical pile foundation construction coefficient of variation to determine the optimal alarm sensitivity parameter; combine effective historical pile foundation construction coefficient of variation and optimal alarm sensitivity parameter to perform adaptive abnormal early warning threshold processing on dynamic baseline value data of pile foundation construction to obtain dynamic construction early warning threshold curve data.

[0071] S4: Collect real-time status data of pile foundation construction equipment; construct a pile foundation monitoring resource topology based on the real-time pile foundation construction equipment status data, and generate pile foundation monitoring resource topology data; construct a pile foundation construction condition prediction model by combining the pile foundation monitoring resource topology data and historical multi-source time series data; use this model to predict short-term construction condition parameters and obtain short-term construction parameter prediction data.

[0072] S5: Calculate the FCR value based on short-term construction parameter prediction data and real-time image features, and combine the dynamic construction early warning threshold curve data to calculate the threshold condition difference and determine the multi-level early warning trigger, forming a multi-level early warning trigger strategy for pile foundations.

[0073] Preferably, S1 includes the following steps:

[0074] S11: Collect historical multi-source pile foundation construction data;

[0075] S12: Perform multi-source format unification and preprocessing on historical multi-source pile foundation construction data to generate standard historical multi-source pile foundation construction data;

[0076] S13: Classify the standard historical multi-source pile foundation construction data according to the historical construction monitoring date to obtain classified historical multi-source pile foundation construction data;

[0077] S14: Perform fast Fourier transform on the classified historical multi-source pile foundation construction data to generate historical pile foundation construction spectrum data;

[0078] S15: Based on historical pile foundation construction spectrum data, perform time-frequency component decomposition of construction parameters to generate time-frequency component data of pile foundation construction parameters.

[0079] In this embodiment of the invention, historical multi-source pile foundation construction data is collected from multiple data sources, including a pile foundation construction monitoring platform database, an intelligent pile driving equipment data acquisition system, a mud circulation monitoring system, a geotechnical engineering monitoring terminal, and historical data and image databases from dedicated working condition sensors. The collected data types include drill rod torque, drilling speed, mud performance parameters (water content, pH value, electrical conductivity), borehole wall displacement, pore water pressure, pile temperature, and corresponding image features for different geological types and pile types.

[0080] The classified historical multi-source pile foundation construction data obtained from S13 are processed using Fast Fourier Transform (FFT) to convert the time-domain construction parameter signals into frequency-domain signals. The amplitude and phase information of different frequency components are analyzed to capture the periodic variation characteristics of the construction parameters. For example, after classifying the weekday construction data of cohesive soil plots, FFT processing is performed to obtain the amplitudes of frequency components such as daily and weekly cycles, corresponding to the periodic fluctuations in intraday construction rhythm and intraweek construction intensity. Time-frequency component decomposition decomposes the original construction data into fluctuation components of different frequencies, including baseline components, daily cycle components, weekly cycle components, and random fluctuation components. Wavelet transform is used to achieve this decomposition. For example, the weekday construction data of cohesive soil plots is decomposed into geotechnical characteristic baseline components, intraday construction rhythm components, and random fluctuation components related to sudden construction conditions.

[0081] Preferably, S15 includes: determining the main fluctuation period (such as daily period, weekly period, seasonal period) of the construction parameters based on spectrum analysis, selecting the appropriate wavelet basis function and decomposition level accordingly, and decomposing the construction parameter sequence into long-term trend, intraday fluctuation, intraweek fluctuation and seasonal fluctuation components to obtain the time-frequency component data of the pile foundation construction parameters.

[0082] In this embodiment of the invention, the main fluctuation period is determined by spectrum analysis, and the construction parameter sequence is decomposed into long-term trend, intraday fluctuation, intraweek fluctuation and seasonal fluctuation components by wavelet transform.

[0083] For example, analysis of weekday spectral data for cohesive soil plots revealed that energy was concentrated in three frequency bands: low frequency (0-0.0417Hz, corresponding to a 24-hour cycle), corresponding to intraday construction rhythm variations; mid frequency (0.125-0.1667Hz, corresponding to a 6-8 hour cycle), corresponding to intraday fluctuations during peak construction periods; and high frequency (above 1Hz), corresponding to instantaneous fluctuations during sudden construction events (such as minor borehole collapse precursors). The frequency ranges of each major band were recorded, and similar analyses were conducted on data from sandy soil, rocky plots, and weekend / holiday data to create a complete historical construction frequency range dataset.

[0084] The approximate coefficients after wavelet decomposition of various data types are extracted and reconstructed into time-domain signals through inverse wavelet transform. These signals serve as long-term trend data for pile foundation construction, reflecting the influence of long-term stable factors such as soil and rock properties and regional construction environment. Detail coefficients at different levels are extracted, corresponding to different periodic fluctuation data: high-frequency detail coefficients reflect instantaneous sudden fluctuations, serving as intraday construction fluctuation data; mid-frequency detail coefficients reflect weekly periodic fluctuations, serving as intraweek construction fluctuation data; and low-frequency detail coefficients reflect seasonal climate influences, serving as seasonal fluctuation data. All of these are reconstructed into time-domain data through inverse transform. The above four types of component data are linked and integrated according to timestamps to form complete time-frequency component data of pile foundation construction parameters, stored as a multidimensional data table, with timestamps as indexes and each component as a field.

[0085] Preferably, S2 includes the following steps:

[0086] S21: Based on the time-frequency component data of pile foundation construction parameters, set the rolling window of construction parameters to obtain the annual trend baseline rolling window, intraday fluctuation baseline rolling window, weekly fluctuation baseline rolling window and seasonal fluctuation baseline rolling window respectively;

[0087] S22: Calculate the long-term trend baseline of pile foundation construction long-term trend data through the annual trend baseline rolling window, and generate long-term trend baseline data;

[0088] S23: Calculate the intraday fluctuation baseline of the pile foundation construction intraday fluctuation data through the intraday fluctuation baseline rolling window, and generate intraday fluctuation baseline data;

[0089] S24: Calculate the weekly fluctuation baseline of the pile foundation construction period fluctuation data through the weekly fluctuation baseline rolling window, and generate the weekly fluctuation baseline data;

[0090] S25: Calculate the seasonal fluctuation baseline of pile foundation construction seasonal fluctuation data through the seasonal fluctuation baseline rolling window, and generate seasonal fluctuation baseline data;

[0091] S26: Perform dynamic baseline value weighting and fusion on long-term trend baseline data, intraday fluctuation baseline data, weekly fluctuation baseline data and seasonal fluctuation baseline data to obtain dynamic baseline value data for pile foundation construction.

[0092] In this embodiment of the invention, a differentiated rolling window is set by combining the periodic characteristics of each time-frequency component and the characteristics of sensor parameters: the annual trend baseline rolling window matches the long-term stability of soil and rock properties and regional construction environment; the intraday fluctuation baseline rolling window is set to 24 hours to capture the rhythm changes of intraday construction periods (morning peak, midday lull, evening peak), wherein the pile temperature window can be refined into 4-hour segments; the weekly fluctuation baseline rolling window is set to 168 hours to match the adjustment pattern of weekly construction plans (high intensity on weekdays, low intensity on weekends); the seasonal fluctuation baseline rolling window is set to 90 days to correspond to the influence cycle of the four seasons on construction, and the window size can be finely adjusted according to the actual construction data patterns.

[0093] Long-term trend baseline values ​​are calculated using the moving average or moving median method through an annual trend baseline rolling window. For example, the average historical parameters for the same geological conditions and borehole depth range are calculated daily as the baseline value. Abnormal samples in geological mineralization are excluded from the mud conductivity baseline. Intraday fluctuation baseline values ​​are calculated using a 24-hour rolling window, taking the average historical parameters for the same construction period within the past 24 hours to capture the periodic changes in intraday construction rhythm. For example, drilling peaks between 8-10 AM daily, corresponding to higher torque baseline values ​​than other times. Weekly fluctuation baseline values ​​are calculated using a 168-hour rolling window, taking the average historical parameters for the same day of the week and the same construction period within the past week to match weekly construction intensity differences. For example, construction intensity between 2-4 PM on weekdays is higher than the same period on weekends. Seasonal fluctuation baseline values ​​are calculated using a 90-day rolling window, taking the average historical parameters for the same date and climatic conditions within the past 90 days to reflect the impact of seasons on construction. For example, the mud moisture content baseline value is higher during the rainy season than during the dry season.

[0094] Based on the influence weight of each component on construction safety (which can be determined through expert experience or statistical analysis of historical data), dynamic weights are assigned to each baseline component and weighted fusion is performed to obtain dynamic baseline values.

[0095] Preferably, S3 includes the following steps:

[0096] S31: Obtain the real-time pile foundation monitoring timestamp;

[0097] S32: Based on the real-time pile foundation monitoring timestamp, filter the valid data in the standard historical multi-source pile foundation construction data, calculate the coefficient of variation, and generate the valid historical construction coefficient of variation of pile foundations;

[0098] S33: Based on the effective historical construction variation coefficient of pile foundation, conduct sensitivity parameter analysis of early warning source to determine the optimal alarm sensitivity parameter;

[0099] S34: Calculate the construction early warning threshold by combining the effective historical construction variation coefficient of the pile foundation with the optimal alarm sensitivity parameter, and apply adaptive threshold constraints to the dynamic baseline value data of the pile foundation construction to obtain the dynamic construction early warning threshold curve data.

[0100] In this embodiment of the invention, the real-time monitoring timestamp of the current pile foundation construction is obtained to filter historical valid data that matches the current working conditions. The filtering is based on geological conditions, calendar dates (seasons, holidays), and borehole depth ranges. Historical construction data from the past 3-5 years with the same dates, similar geological conditions, the same borehole depth range, and climate type are selected. Abnormal data exceeding ±3 times the standard deviation of the mean are removed to ensure the validity and representativeness of the data. The stage-specific differences in construction can be fully reflected by the parameter sequence changing with time and borehole depth, without the need to introduce additional fuzzy conditions for determining the construction stage. For the filtered valid data, the coefficient of variation for each data set is calculated using the formula (the coefficient of variation is the ratio of the standard deviation to the mean) to quantify the dispersion of construction parameters and reflect the stability of parameters under different working conditions. For example, the coefficient of variation for sandy soil is usually higher than that for cohesive soil, corresponding to more significant fluctuations in construction parameters. The coefficient of variation for mud conductivity needs to be separately matched to geological differences.

[0101] The sensitivity parameter range and step size are set, and the initial sensitivity parameter is set based on the coefficient of variation range (low coefficient of variation < 0.1, medium coefficient of variation 0.1-0.3, high coefficient of variation > 0.3). The higher the coefficient of variation, the lower the initial sensitivity parameter to avoid false alarms. The historical early warning database for pile foundations is accessed to extract real abnormal alarms (related to accident records such as borehole collapse, stuck drill, and mud instability) and false alarm records. Simultaneously, corresponding historical image features (texture features, color distribution, depth features), FCR values, and LSC values ​​are associated. Through backtracking tests, recall, precision, and F1 scores under different sensitivity parameters are calculated. The parameter with the highest F1 score is selected as the optimal alarm sensitivity parameter to balance the risks of missed and false alarms.

[0102] Calculate the upper and lower limits of the warning threshold using the formula:

[0103] Upper limit = dynamic baseline value + (optimal sensitivity parameter × coefficient of variation);

[0104] Lower limit = Dynamic baseline value - (Optimal sensitivity parameter × coefficient of variation);

[0105] At the same time, threshold constraint ranges are set based on engineering specifications and safety thresholds to avoid warning failures caused by thresholds that are too high or too low. Ultimately, threshold curve data that is dynamically adjusted according to operating conditions is generated, and independent threshold constraint rules are set for sensitive parameters.

[0106] Preferably, S32 includes the following steps:

[0107] S321: Determine the effective historical construction window based on the real-time pile foundation monitoring timestamp to obtain effective historical construction window data;

[0108] S322: Filter standard historical multi-source pile foundation construction data through effective historical construction window data to obtain effective historical pile foundation construction data;

[0109] S323: Calculate the historical construction average based on the effective historical construction data of pile foundations, and generate effective historical construction average data;

[0110] S324: Calculate the historical construction standard deviation based on the effective historical construction data of pile foundations, and generate effective historical standard deviation data;

[0111] S325: Calculate the coefficient of variation by combining the effective historical construction mean data and the effective historical standard deviation data to generate the effective historical construction coefficient of variation for pile foundations.

[0112] In this embodiment of the invention, the date, time period, sensor deployment location, and current borehole depth information contained in the real-time timestamp are extracted. A valid historical construction window (e.g., a time period of 1 hour before and after the current period) is defined. Window data from the same day, borehole depth range, and sensor location within the past 3 years are selected. The window width can be adjusted according to the fluctuation rate of construction parameters (a narrower window for faster fluctuation rates and a wider window for slower fluctuation rates; for example, a 30-minute window for torque parameters and a 2-hour window for temperature parameters). Standard historical data is traversed, and data whose timestamps fall within the valid window and match geological and climatic conditions are filtered. After removing outliers, a valid historical construction dataset for pile foundations is formed and stored in a list format. Each data entry includes a timestamp, construction parameter value, associated working condition label, sensor location, borehole depth range, and image features (texture, color, and depth feature quantification values).

[0113] The mean is calculated for the data within each valid window. A multi-window mean-plus-average method can be used to optimize the results, combined with moving average smoothing to improve the stability of the mean data. For sensitive parameters, a weighted average method is used to increase the weight of samples under normal operating conditions. Similarly, the standard deviation is calculated for the data within each valid window. Multi-window standard deviation-plus-average and moving average smoothing are used to reduce the impact of random fluctuations on the standard deviation. The coefficient of variation is calculated window-by-window using the formula to obtain the coefficient of variation for each window, or a comprehensive coefficient of variation can be calculated using the overall mean and overall standard deviation. Simultaneously, the coefficient of variation for sensitive parameters is calculated separately to provide data support for subsequent sensitive parameter analysis.

[0114] Preferably, S33 includes the following steps:

[0115] S331: Based on the analysis of the coefficient of variation distribution characteristics of effective pile foundation historical construction, the initial sensitivity parameter range is set, and the initial sensitivity parameter range data is obtained;

[0116] S332: Obtain configuration data for early warning sources of pile foundation construction load;

[0117] S333: Based on the configuration data of the pile foundation construction load early warning source, access the early warning source, read historical early warning records, and obtain the original pile foundation early warning record data;

[0118] S334: Based on the original pile foundation early warning record data, filter out the real alarm records to obtain the real alarm record data of the pile foundation;

[0119] S335: Based on the original pile foundation early warning record data, filter the false alarm records to obtain the pile foundation false alarm record data;

[0120] S336: According to the preset sensitivity traversal step size, the initial sensitivity parameter range data is traversed in a loop to obtain the sensitivity parameter traversal sequence;

[0121] S337: Calculate the alarm recall rate of the actual alarm record data and the false alarm record data of the pile foundation based on the sensitivity parameter traversal sequence to obtain the alarm recall rate data of the pile foundation;

[0122] S338: Calculate the alarm accuracy rate of the actual alarm record data and the false alarm record data of the pile foundation based on the sensitivity parameter traversal sequence to obtain the alarm accuracy rate data of the pile foundation;

[0123] S339: Combine the pile foundation alarm accuracy data and the pile foundation alarm recall data to calculate the early warning performance index, select the optimal sensitivity parameter, and obtain the optimal alarm sensitivity parameter.

[0124] In this embodiment of the invention, statistical analysis is performed on the effective historical construction variation coefficients of pile foundations, calculating indicators such as minimum, maximum, mean, and median values. Histograms or box plots are then drawn to observe distribution characteristics. For example, most variation coefficients are concentrated in the 0.1-0.3 range. Based on this, an initial sensitivity parameter range is set, and parameter sub-ranges are divided according to the variation coefficient range. Simultaneously, narrower parameter sub-ranges are set for sensitive parameters (drill rod torque, mud pH value) to achieve differentiated initial settings. Configuration data of the pile foundation construction early warning source is obtained, including database connection information (address, username, password), early warning system parameters (alarm level classification, information transmission method), sensor and camera configuration information, etc., and access to the early warning source is achieved through API interface or database connection.

[0125] Read raw early warning records from the past 1-3 years, including early warning time, construction location, geological type, early warning parameters, early warning level, sensor location, borehole depth range, corresponding image features (texture features, color distribution, depth features), FCR value, LSC value, etc. Combined with historical construction accident records and technical personnel verification reports, early warning records corresponding to actual abnormal events (borehole collapse, stuck drill, etc.) are selected as actual pile foundation alarm records; early warning records without corresponding actual abnormal events, verified as being caused by instantaneous parameter fluctuations or image interference, are selected as false alarm records. The initial sensitivity parameter range is traversed at a preset step size to generate a complete sensitivity parameter traversal sequence.

[0126] For each traversal parameter, recall (the ratio of actual alarms detected to the total number of actual alarms) is calculated to quantify the parameter's coverage of actual anomalies; simultaneously, precision (the ratio of actual alarms to the total number of alerts) is calculated to quantify the accuracy of the alerts. The F1 score (harmonic mean) is used as the core performance indicator, and the formula is as follows: Alternatively, the F2 value (emphasizing recall) can be selected according to requirements. All parameters are traversed and the corresponding performance indicators are calculated. The parameter with the best early warning performance among sensitive parameters is selected as the optimal alarm sensitivity parameter.

[0127] The optimal alarm sensitivity parameter is a dimensionless coefficient with a value range of [0.1, 1.0]. Physically, it acts as an adjustment factor for the tightness of anomaly detection, quantifying the maximum reasonable range of deviation of construction parameters from the dynamic baseline. It directly determines the distance between the warning threshold and the dynamic baseline. A larger value results in more sensitive warning detection (closer distance between the threshold and the baseline), making it easier to detect minor anomalies, but may increase the false alarm rate. A smaller value results in more lenient warning detection (farther distance between the threshold and the baseline), reducing false alarms, but may miss minor anomalies. The calculation method is: dynamic baseline value ± (sensitivity parameter × coefficient of variation).

[0128] As an embodiment of the present invention, see [reference]. Figure 3 As shown, in this embodiment, S4 includes:

[0129] S41: Collect real-time status data of pile foundation construction equipment and synchronous image features;

[0130] In this embodiment of the invention, real-time status data and synchronous image data of pile foundation construction equipment are collected from multiple data sources, including SCADA monitoring systems, intelligent sensors and equipment controllers, dedicated working condition sensors, and cameras. Specifically, the data collected by the dedicated working condition sensors includes: real-time pile temperature, pile end mud pH value, and moisture content collected by the pile body working condition sensor; drilling status sensor collected drill end mud pH value, conductivity, and drill rod torque; and the camera periodically captures images of the mud surface at the borehole opening. Multi-dimensional image features, including texture features and color distribution, are extracted using image processing techniques, and depth features can be extracted using a pre-trained CNN model if necessary.

[0131] S42: Construct a pile foundation monitoring resource topology based on real-time pile foundation construction equipment status data, and generate pile foundation monitoring resource topology data. In this embodiment of the invention, based on the real-time data collected in S41, a pile foundation monitoring resource topology is constructed. A graph data structure (adjacency matrix or adjacency list) is used to describe the connection relationships and data transmission paths of each monitoring unit, construction equipment, sensor, camera, and control terminal. In the topology graph, nodes represent core construction and monitoring components such as pile driving equipment, mud circulation system, borehole wall monitoring points, pile condition sensors, drilling status sensors, cameras, and control terminals. Edges represent equipment mechanical connections and data transmission links. Both nodes and edges are accompanied by attribute information (such as equipment model, sensor type, monitoring range, transmission rate, geological number, image acquisition angle, and construction stage adaptability). The spatial correlation features of each monitoring parameter are extracted through the topology structure, clarifying the coupling relationship of core parameters such as drill rod torque, mud performance, and borehole wall displacement, providing accurate support for the fusion of input features for the time series model.

[0132] S43: Construct historical multi-source time-series input samples and associate them with topological spatial features;

[0133] In this embodiment of the invention, the time-series parameters (drill rod torque, mud moisture content, borehole wall displacement, pile temperature, mud conductivity, and mud pH) corresponding to the core monitoring nodes are determined based on the topology. Combined with synchronously acquired multi-dimensional image features (texture features, color distribution, and depth features), a time-series input sample is constructed using the sliding window method. The high-frequency data at the second level is first aggregated using a 1-minute mean. Then, the sliding window length is set to the past 24 hours (144 time steps at a 10-minute granularity), and the prediction step size is the future 24 hours (144 time steps). Each sample contains a sequence of time-series parameters within the window, corresponding time-step image features, and a topological spatial feature vector. The topological spatial feature vector is generated through the topological graph adjacency matrix and node attribute transformation. After being concatenated with the time-series features and image features, it forms a complete input feature set for model training and prediction.

[0134] S44: Construct an LSTM model based on multi-source temporal features to obtain a prediction model for pile foundation construction conditions to be trained;

[0135] In this embodiment of the invention, a Long Short-Term Memory (LSTM) network is selected as the core prediction model to meet the need for capturing long-term dependencies in multi-source time series data. The core task of the model is to input historical multi-source time series data (including time series parameters, image features, and topological spatial features) and output the predicted values ​​of core construction parameters for a future period of time.

[0136] The model structure is designed as follows: The input layer receives the concatenated feature vector (dimension is time step × number of features, the number of features includes time-series parameters, image features, and topological features); the hidden layer consists of two LSTM layers, the first layer containing 64 neurons and the second layer containing 32 neurons, with a Dropout layer used to suppress overfitting and capture the long-term and short-term dependencies of the time-series data; the output layer is a fully connected layer, with the number of neurons corresponding to the prediction step size × the number of core construction parameters (core parameters include drill rod torque, mud moisture content, borehole wall displacement, pile temperature, and mud pH value), and the activation function is a Linear function, directly outputting the predicted parameter values ​​for each future time step. The model is compiled using the Adam optimizer, and the loss function is the mean squared error, suitable for continuous value time-series prediction tasks.

[0137] S45: Time-series sample processing is performed on standard historical multi-source pile foundation construction data to train an LSTM model and obtain a pile foundation construction condition prediction model. In this embodiment of the invention, standard historical multi-source data is used to construct time-series training samples according to the sliding window rule in S43. Date type, geological type, climate data, and borehole depth range are extracted as auxiliary features and incorporated into the input. The data is divided into training set and test set according to the ratio, and all features are normalized. Batch training mode is adopted, and iterative training is performed. The prediction accuracy is verified through the test set in each round. Training stops when the MSE of the test set is lower than the preset threshold and there is no decrease for a set number of consecutive rounds.

[0138] S46: Use the pile foundation construction condition prediction model to predict short-term construction parameters and obtain short-term construction parameter prediction data;

[0139] In this embodiment of the invention, the real-time time-series input samples constructed by S43 are input into the trained LSTM model to carry out short-term construction parameter prediction for the next 24 hours. The predicted values ​​of core construction parameters (drill rod torque, mud moisture content, borehole wall displacement, pile temperature, and mud pH value) are output at a time step of 10 minutes and stored in time series form to provide core data support for subsequent early warning threshold comparison, FCR value calculation and multi-level early warning judgment.

[0140] In this embodiment, S5 includes:

[0141] S51: Extract the prediction time point of short-term construction parameter prediction data and simultaneously collect real-time image features of the corresponding time period;

[0142] In this embodiment of the invention, short-term construction parameter prediction data is stored in time series format. The timestamps corresponding to each prediction parameter are extracted as prediction time point data to form a timestamp sequence, ensuring consistency with the time dimension of the dynamic threshold curve and real-time image features. Simultaneously, a camera is invoked to collect real-time images corresponding to the prediction time period, extracting multi-dimensional image features such as texture features, color distribution, and depth features to provide basic data for FCR value calculation and anomaly identification.

[0143] S52: Calculate the FCR value based on short-term construction parameter prediction data and real-time image features to complete the construction of early warning auxiliary indicators; In this embodiment of the invention, the definition of core indicators is first clarified:

[0144] The FCR value (Feature Co-anomaly Coefficient) is an auxiliary early warning indicator that quantifies the degree of coordination between construction parameter prediction anomalies and image feature anomalies, and is used to make up for the limitations of single parameter early warning.

[0145] The LSC value (stability characteristic coefficient of pile body) is a core stability index calculated based on pile body working parameters and image depth features, directly reflecting the safety status of the borehole wall and pile body structure.

[0146] The FCR value is calculated based on the correlation between short-term construction parameter predictions and real-time image features (texture features, color distribution, and depth features). The formula is: FCR = α × (predicted torque deviation from baseline) + β × (image texture complexity deviation from standard value) + γ × (color distribution anomaly coefficient) + δ × (depth feature matching deviation), where α, β, γ, and δ are weighting coefficients (calibrated by expert experience and historical data, with a sum of 1, where α = 0.3, β = 0.25, γ = 0.2, and δ = 0.25), and each deviation is normalized to the [0,1] interval; the predicted torque deviation from baseline = (predicted torque value - dynamic baseline torque value) / dynamic baseline torque value; the image texture complexity deviation from standard value is calculated based on the entropy value extracted from the gray-level co-occurrence matrix (GLCM) as a quantitative indicator of texture complexity. The larger the entropy value, the more disordered the texture. ;

[0147] Method for calculating the color distribution anomaly coefficient: Extract the standard deviation of the RGB three channels of the image. , , Under normal working conditions, the color distribution is uniform (small standard deviation), but under abnormal conditions, the standard deviation increases due to factors such as the sand content and air bubbles in the mud. The method for calculating the depth feature matching deviation is as follows: the image depth feature vector (dimension 256) is extracted by the pre-trained ResNet-18 model, and the average cosine similarity Sim between the real-time depth feature vector and the normal working condition feature library is calculated. The depth feature matching deviation = 1 − Sim (the lower the similarity, the greater the deviation).

[0148] The LSC value is calculated as follows: LSC = (ratio of predicted borehole wall displacement to safety threshold) × 0.4 + (mud stability index) × 0.3 + (image depth feature safety matching degree) × 0.3, where the mud stability index is calculated by combining mud pH value, water content, and conductivity.

[0149] The method for calculating the mud stability index is as follows: It integrates three parameters—mud pH, water content, and electrical conductivity—and uses a weighted summation followed by normalization. , , , These are the weighting coefficients.

[0150] The image depth feature safety matching degree is the cosine similarity between the real-time depth feature and the normal working condition feature library. The LSC value ranges from [0,2]. The higher the value, the worse the stability of the pile.

[0151] S53: Calculate the extent of parameter exceedance by matching dynamic early warning thresholds with predicted time point data; .

[0152] In this embodiment of the invention, the prediction time points are iterated one by one, and the upper and lower limits of the threshold values ​​of each parameter at the corresponding time point are matched in the dynamic threshold curve. If the time points are not perfectly aligned, the corresponding threshold is calculated using interpolation. The over-limit amplitude of each parameter at each time point is calculated according to the formula (over-limit amplitude = (predicted value - threshold) / threshold × 100%). When the predicted value is higher than the upper limit, it is a positive amplitude; when it is lower than the lower limit, it is a negative amplitude.

[0153] S54: Combine the over-limit amplitude, FCR value and LSC value to classify the warning level and generate a multi-level warning triggering strategy;

[0154] In this embodiment of the invention, a four-level early warning system is constructed by combining the parameter over-limit amplitude, FCR value, image features, and LSC value. The division of each level and the execution strategy are as follows:

[0155] Level 4 warning (minor anomaly): Parameter exceeds the limit by 3%-5% or -3%≤exceedance <-1%, and FCR < 0.3 (standard threshold), image features have no obvious abnormalities (texture features and color distribution are within the normal operating range), and LSC < 0.5 (preset threshold).

[0156] Strategy: Increase the frequency of parameter monitoring (once every 5 minutes), and simultaneously increase the frequency of image acquisition (1 frame every 2 minutes). Focus on monitoring the changing trends of image texture features and color distribution, record data and changes in image features, without interrupting construction, and regularly push monitoring reports to the control terminal.

[0157] Level 3 warning (general anomaly): Parameter exceeds the limit by 5%-10% or -8%≤exceedance <-3%, or FCR≥0.3 and <0.6 (1.2 times the standard threshold), slight abnormalities in image features (such as texture complexity fluctuation of 10%-15%, local color shift), LSC≥0.5 and <1.0 (upgraded threshold).

[0158] Strategy: Notify the construction control team and technical personnel to conduct on-site verification, and combine image features (such as mud texture roughness and abnormal color patches) to determine the sand content and air bubble content of the mud. Reduce drilling speed, adjust mud properties (such as adjusting pH value and water content), increase monitoring frequency, and submit a verification report and image feature analysis results every 10 minutes.

[0159] Level 2 warning (serious anomaly): Parameter exceeds the limit by more than 10% or less than -8%, or FCR ≥ 0.6, obvious abnormalities in image features (such as texture disorder, large color shift, identifiable bubble clusters), LSC ≥ 1.0.

[0160] Strategy: Immediately send a level-two warning signal to the central control terminal, suspend the drilling process of the pile foundation drilling operation, initiate the emergency investigation process, further confirm the signs of borehole collapse and the degree of mud instability by combining image depth features, organize the technical team to investigate the risk of borehole collapse and stuck drill, and simultaneously analyze the correlation between sensor data and image features to clarify the cause of the anomaly.

[0161] Level 1 Warning (Extreme Anomaly): If, after Level 2 warning investigation, the LSC continues to rise above 1.5 (danger threshold), or if the image shows obvious signs of collapse, a large number of bubble clusters accompanied by a sudden change in mud color, and a sudden increase in torque (exceeding the limit by more than 20%), the risk of borehole wall instability is determined to be extremely high.

[0162] Strategy: Upgrade the warning level to Level 1, halt all pile foundation construction operations, evacuate unrelated personnel from the site, activate the emergency response plan for the collapsed borehole, report to the construction management department simultaneously, and record the entire handling process and image feature changes to provide a basis for subsequent review and analysis.

[0163] The above-mentioned early warning strategy can be dynamically optimized by combining engineering specifications, field experience and historical alarm data. The thresholds, FCR coefficients α / β / γ / δ and LSC coefficients of each early warning level can be calibrated specifically according to different geological types and pile types. The criteria for judging the anomalies of image features need to be based on a benchmark library constructed from historical normal working condition samples.

[0164] The beneficial effects of this embodiment are as follows: by multi-source data fusion, time-frequency component decomposition and dynamic baseline weighting, combined with LSTM prediction and multi-level early warning strategy, accurate prediction of construction parameters and collaborative identification of anomalies are achieved, improving the real-time performance, accuracy and adaptability of early warning, reducing the risk of false alarms and missed alarms, and supporting the refined management and control of pile foundation construction safety.

[0165] All formulas in this invention are dimensionless and calculated numerically. The preset parameters in the formulas can be set by those skilled in the art according to the actual situation.

[0166] The weighting coefficients of this invention are used to measure the degree of influence of different factors or variables on a certain outcome or decision. The weighting coefficient is defined as the numerical value assigned to each factor when comparing and evaluating multiple factors, reflecting their importance or priority. These weighting coefficients can be determined according to specific circumstances and needs, and are usually jointly formulated and confirmed by professionals or relevant stakeholders. By reasonably setting the weighting coefficients, programs or systems can be helped to make decisions or predictions more accurately.

[0167] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0168] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An intelligent identification and early warning system for abnormal working conditions in pile foundation construction, characterized in that, include: Historical data processing module: used to acquire standard historical multi-source pile foundation construction data, classify it by date, and generate time-frequency component data of pile foundation construction parameters through FFT and wavelet transform; Dynamic baseline generation module: used to set a differentiated rolling window based on time-frequency component data, and obtain dynamic baseline value data for pile foundation construction by weighted fusion; Threshold calculation module: used to filter valid historical data and calculate the coefficient of variation, determine the optimal alarm sensitivity parameters, and generate dynamic construction early warning threshold curve data; The working condition prediction module is used to collect real-time data and construct the topology of pile foundation monitoring resources. It performs short-term construction parameter prediction based on the LSTM model and outputs short-term construction parameter prediction data. Early warning triggering module: used to calculate the characteristic co-abnormality coefficient FCR value and the pile stability characteristic coefficient LSC value, and combine the dynamic early warning threshold and the over-limit amplitude to divide the early warning level and generate a multi-level early warning triggering strategy for pile foundation.

2. An intelligent identification and early warning method for abnormal working conditions in pile foundation construction, used to implement the intelligent identification and early warning system for abnormal working conditions in pile foundation construction as described in claim 1, characterized in that, The method includes the following steps: S1: Obtain standard historical multi-source pile foundation construction data; classify the standard historical multi-source pile foundation construction data according to the historical construction monitoring date, perform time-frequency component decomposition of construction parameters, and generate time-frequency component data of pile foundation construction parameters; S2: Based on the time-frequency component data of the pile foundation construction parameters, set a rolling window for construction parameters, perform dynamic baseline value weighting and fusion, and obtain dynamic baseline value data for pile foundation construction. S3: Perform effective data screening and coefficient of variation calculation on the standard historical multi-source pile foundation construction data to generate effective historical pile foundation construction coefficient of variation; conduct sensitivity parameter analysis of early warning sources based on the effective historical pile foundation construction coefficient of variation to determine the optimal alarm sensitivity parameter; combine the effective historical pile foundation construction coefficient of variation and the optimal alarm sensitivity parameter to perform adaptive abnormal early warning threshold processing on the dynamic baseline value data of pile foundation construction to obtain dynamic construction early warning threshold curve data; S4: Collect real-time status data of pile foundation construction equipment; construct a pile foundation monitoring resource topology based on the real-time pile foundation construction equipment status data, and generate pile foundation monitoring resource topology data; construct a pile foundation construction condition prediction model by combining the pile foundation monitoring resource topology data and historical multi-source time series data; use the model to predict short-term construction condition parameters and obtain short-term construction parameter prediction data. S5: Calculate the FCR value based on the short-term construction parameter prediction data and real-time image features, and combine the dynamic construction early warning threshold curve data to calculate the threshold condition difference and determine the multi-level early warning trigger, thus forming a multi-level early warning trigger strategy for pile foundations.

3. The intelligent identification and early warning method for abnormal working conditions in pile foundation construction according to claim 2, characterized in that, Step S1 includes the following sub-steps: S11: Collect historical multi-source pile foundation construction data. The historical multi-source pile foundation construction data comes from various construction monitoring related databases and sensor historical records. The data types include construction parameters and corresponding image features for different geological types and different pile types. S12: Perform multi-source data source format unification and preprocessing on the historical multi-source pile foundation construction data to generate the standard historical multi-source pile foundation construction data; S13: Classify the standard historical multi-source pile foundation construction data according to the historical construction monitoring date to obtain classified historical multi-source pile foundation construction data; S14: Perform frequency domain transformation processing on the classified historical multi-source pile foundation construction data to generate historical pile foundation construction spectrum data; S15: Based on the historical pile foundation construction spectrum data, perform time-frequency component decomposition of construction parameters to generate the time-frequency component data of the pile foundation construction parameters; the time-frequency component decomposition adopts wavelet transform, and the main fluctuation period of construction parameters is determined based on spectrum analysis during the decomposition process, and the appropriate wavelet basis function and decomposition level are selected accordingly.

4. The intelligent identification and early warning method for abnormal working conditions in pile foundation construction according to claim 2, characterized in that, Step S2 includes the following sub-steps: S21: Based on the time-frequency component data of the pile foundation construction parameters, set the construction parameter rolling window to obtain the rolling window corresponding to each time-frequency component; S22: Calculate the baseline data of the corresponding time-frequency components through each rolling window; S23: Based on the influence weight of each component on construction safety, assign dynamic weights to each baseline component and perform weighted fusion to obtain the dynamic baseline value data of the pile foundation construction; the weights are determined through expert experience or statistical analysis of historical data.

5. The intelligent identification and early warning method for abnormal working conditions in pile foundation construction according to claim 2, characterized in that, Step S3 includes the following sub-steps: S31: Obtain the real-time pile foundation monitoring timestamp; S32: Based on the real-time pile foundation monitoring timestamp, filter the valid data in the standard historical multi-source pile foundation construction data, calculate the coefficient of variation, and generate the valid historical construction coefficient of variation of the pile foundation. S33: Based on the effective historical construction variation coefficient of the pile foundation, conduct sensitivity parameter analysis of the early warning source to determine the optimal alarm sensitivity parameter; during the analysis, set the initial sensitivity parameter in combination with the variation coefficient range, calculate the alarm performance index under different sensitivity parameters through backtracking test, and select the parameter with the best performance as the optimal alarm sensitivity parameter; S34: Calculate the construction early warning threshold by combining the effective historical construction variation coefficient of the pile foundation with the optimal alarm sensitivity parameter, set the threshold constraint range based on engineering specifications and safety thresholds, and apply adaptive threshold constraints to the dynamic baseline value data of the pile foundation construction to obtain the dynamic construction early warning threshold curve data.

6. The intelligent identification and early warning method for abnormal working conditions in pile foundation construction according to claim 5, characterized in that, Step S32 includes the following sub-steps: S321: Determine the historical valid construction window based on the real-time pile foundation monitoring timestamp, and obtain the valid historical construction window data; S322: Filter the standard historical multi-source pile foundation construction data through the effective historical construction window data, and obtain the effective historical construction data of pile foundations after removing outliers; S323: Calculate the mean and standard deviation based on the effective historical construction data of the pile foundation; S324: Calculate the coefficient of variation by combining the mean and standard deviation to generate the effective historical construction coefficient of variation of the pile foundation.

7. The intelligent identification and early warning method for abnormal working conditions in pile foundation construction according to claim 5, characterized in that, In step S33, statistical analysis is performed on the effective historical construction variation coefficient of the pile foundation to observe its distribution characteristics. The parameter sub-ranges are divided according to the variation coefficient interval, and a narrower parameter sub-range is set for sensitive parameters. During the backtest, real abnormal alarm and false alarm records are extracted, the initial sensitivity parameter range is traversed, and the corresponding alarm performance index is calculated. The parameter with the best early warning performance of the sensitive parameter is selected first.

8. The intelligent identification and early warning method for abnormal working conditions in pile foundation construction according to claim 2, characterized in that, Step S4 includes the following sub-steps: S41: Collect real-time status data of pile foundation construction equipment and synchronous image features; the image features are extracted using image processing technology; S42: Construct a pile foundation monitoring resource topology based on the real-time pile foundation construction equipment status data, use a graph data structure to describe the connection relationship and data transmission path of each component, and generate the pile foundation monitoring resource topology data; S43: Based on the topology of the pile foundation monitoring resources, determine the time series parameters corresponding to the core monitoring nodes, and combine the synchronous image features and topological spatial features to construct historical multi-source time series input samples using the sliding window method; S44: An LSTM model is constructed based on multi-source temporal features as a prediction model for pile foundation construction conditions to be trained; the model contains an input layer, multiple LSTM hidden layers and a fully connected output layer, and regularization is used to suppress overfitting. It is compiled through an adapter optimizer, and the loss function is selected to be adapted to the type of continuous value prediction. S45: Perform time-series sample processing on the standard historical multi-source pile foundation construction data, divide the training set and test set and normalize them, iteratively train the LSTM model to the preset accuracy, and obtain the pile foundation construction condition prediction model. S46: Use the pile foundation construction condition prediction model to predict short-term construction parameters and obtain the short-term construction parameter prediction data.

9. The intelligent identification and early warning method for abnormal working conditions in pile foundation construction according to claim 2, characterized in that, In step S5, the feature co-anomaly coefficient FCR and the pile stability feature coefficient LSC are calculated. A multi-level early warning system is constructed by combining the parameter over-limit amplitude, FCR value, image features and LSC value, and each level corresponds to a differentiated execution strategy.

Images