Micro pipe jacking construction regulation and control method and system based on multi-source data fusion

Abstract

The invention provides a micro pipe jacking construction regulation and control method based on multi-source data fusion, and the method comprises the steps: obtaining the operation parameters, geological parameters and environmental data of pipe jacking equipment through a multi-source sensor, and generating a multi-source data set; performing space-time alignment and data cleaning on the multi-source data set to obtain a standardized data set; extracting features from the standardized data set through a neural network model, classifying the features, and outputting a dynamic geological classification result; identifying a jacking resistance grade based on the dynamic geological classification result, and determining equipment regulation and control parameters through an optimization algorithm; adjusting the equipment regulation and control parameters according to the real-time monitoring data, and matching a slurry formula scheme; and updating and optimizing regulation and control parameters through a prediction error triggering model, and adjusting a decision threshold according to a segmented evaluation result.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular, to a micro pipe jacking construction control method and system based on multi-source data fusion. Background Art

[0002] With the acceleration of urbanization and the complication of underground space development, this technology has become the key to solving the needs of narrow spaces and high-precision construction. However, existing control methods mostly rely on manual experience or single data sources, making it difficult to cope with complex and changeable stratum conditions and construction environments, resulting in insufficient construction accuracy, increased equipment wear, and even increased engineering risks. These limitations make traditional methods often show insufficient adaptability when facing diverse geological conditions such as sand layers, clay layers or composite strata, and cannot achieve dynamic optimization and intelligent decision-making. In this context, the core challenges faced by micro pipe jacking construction control mainly focus on how to effectively integrate multi-source data, accurately identify geological conditions, and adaptively adjust control parameters. Due to the difficulty in unifying the heterogeneity and real-time nature of multi-source data, existing systems are difficult to build accurate working condition recognition models, which in turn affects the dynamic matching of jacking force, rotation speed, and mud parameters. In addition, the optimization of control strategies lacks the ability of continuous learning, resulting in difficulty in getting rid of experience dependence in complex strata, and limited decision-making efficiency and adaptability. These technical problems not only increase the uncertainty of construction but also limit the application potential of micro pipe jacking technology in a wider range of scenarios. Therefore, how to establish an intelligent working condition recognition model based on multi-source data fusion and dynamically optimize control parameters through adaptive algorithms has become the key issue to improve the adaptability and efficiency of micro pipe jacking construction. Solving this problem requires a complete chain breakthrough from data integration to decision optimization to meet the complex requirements of different stratum conditions and construction stages. Summary of the Invention

[0003] The present invention provides a micro pipe jacking construction control method based on multi-source data fusion, mainly including:

[0004] Obtaining operation parameters, geological parameters, and environmental data of the pipe jacking equipment through multi-source sensors to generate a multi-source data set; performing spatio-temporal alignment and data cleaning on the multi-source data set to obtain a standardized data set; extracting features from the standardized data set through a neural network model and classifying them to output a dynamic geological classification result; identifying the jacking resistance level based on the dynamic geological classification result, and determining equipment control parameters through an optimization algorithm; adjusting the equipment control parameters according to real-time monitoring data and matching the mud formulation plan; triggering model update and optimizing control parameters through prediction error, and adjusting the decision threshold according to the segmented evaluation result.

[0005] Furthermore, the operation parameters, geological parameters, and environmental data of the pipe jacking equipment are obtained through multi-source sensors to generate a multi-source dataset, including: obtaining the operation parameters of the equipment through the inclinometer, earth pressure gauge, and torque meter installed on the pipe jacking equipment; obtaining the geotechnical mechanical parameters through a geological detector; obtaining the groundwater level and surface vibration data through an environmental monitoring device; obtaining the mud viscosity parameters through a rheometer, and generating a formation permeability curve in combination with the data of a permeability tester; integrating the operation parameters of the equipment, the geotechnical mechanical parameters, and the groundwater level and surface vibration data into a multi-source heterogeneous dataset.

[0006] Furthermore, the multi-source dataset is subjected to spatio-temporal alignment and data cleaning to obtain a standardized dataset, including: performing spatio-temporal alignment processing on the multi-source dataset using a distributed Kalman filtering algorithm; removing the abnormal data in the multi-source dataset through an outlier rejection algorithm based on the 3σ principle;

[0007] filling the missing data in the multi-source dataset through a KNN missing value imputation algorithm; generating a standardized dataset with a unified timestamp.

[0008] Furthermore, features are extracted from the standardized dataset through a neural network model and classified to output a dynamic geological classification result, including: inputting the standardized dataset into a convolutional recurrent neural network to extract geological spatial distribution features, equipment power spectrum features, and environmental time-varying features; establishing a feature correlation matrix for the extracted features through a fuzzy C-means clustering algorithm; outputting a dynamic geological classification result including sand layers, clay layers, and composite strata according to the feature correlation matrix.

[0009] Furthermore, the jacking resistance level is identified based on the dynamic geological classification result, and the equipment control parameters are determined through an optimization algorithm, including: inputting the dynamic geological classification result into a pre-trained working condition identification model, and the working condition identification model is constructed using a bidirectional LSTM network containing an attention mechanism; outputting a jacking resistance level label according to the comparison between the geotechnical shear strength eigenvalue and the dynamic decision threshold; calling the historical construction database based on the jacking resistance level label, and calculating the Pareto optimal solution set of the jacking force and the rotation speed using an improved particle swarm algorithm with equipment power constraints and jacking speed constraints.

[0010] Furthermore, the equipment control parameters are adjusted according to the real-time monitoring data and the mud formula scheme is matched, including: real-time monitoring of the jacking force deviation and the rotation speed volatility; if the ratio of the jacking force deviation to the rotation speed volatility exceeds the set tolerance, calculating the compensation increment through a gradient descent optimizer; retrieving the formula scheme with the highest matching degree from the pre-established mud formula knowledge base according to the compensated jacking force combined with the mud viscosity-formation permeability curve; updating the knowledge base parameters based on the implementation effect of the mud formula.

[0011] Further, the triggering of model update and optimization of control parameters by prediction error includes: dividing the real-time construction data stream into sample blocks according to a time window; if the prediction error rate of the working condition recognition model on three consecutive sample blocks exceeds a preset model error threshold, the elastic weight consolidation algorithm is used to update the network weights; injecting the output of the updated model as prior knowledge into the feature association matrix, and generating an iteratively optimized set of control parameters through heterogeneous data fusion.

[0012] Further, the adjustment of the formation disturbance threshold and the equipment decision threshold according to the segmented evaluation results includes: when the jacking distance reaches the segmented threshold, obtaining the soil stress change amount and the weight coefficient of each monitoring point; calculating the formation disturbance coefficient according to the soil stress change amount, the weight coefficient and the average jacking force; dynamically adjusting the decision threshold and the tolerance ratio parameter of the subsequent stage according to the formation disturbance coefficient; the set tolerance is calculated according to the formation disturbance coefficient of the previous construction section.

[0013] The present invention also provides a micro-tunnel jacking construction control system for implementing any one of the methods described in claims 1-8, characterized by including:

[0014] A multi-source data acquisition module, configured to obtain the operation parameters, geological parameters and environmental data of the pipe jacking equipment through multi-source sensors, and generate a multi-source data set;

[0015] A data preprocessing module, configured to perform spatio-temporal alignment and data cleaning on the multi-source data set to obtain a standardized data set;

[0016] A dynamic geological classification module, configured to extract features from the standardized data set through a neural network model and perform classification, and output a dynamic geological classification result;

[0017] A resistance control module, configured to identify the jacking resistance level based on the dynamic geological classification result, and determine the equipment control parameters through an optimization algorithm;

[0018] A real-time decision module, configured to adjust the equipment control parameters according to real-time monitoring data and match the mud formula scheme;

[0019] A model iteration module, configured to trigger model update and optimize control parameters through prediction error;

[0020] A threshold optimization module, configured to adjust the decision threshold according to the segmented evaluation result.

[0021] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0022] The present invention discloses a micro - pipe - jacking construction regulation method based on multi - source data fusion. The method collects the operation parameters of the pipe - jacking equipment, geotechnical mechanics parameters, and environmental data through multi - source sensors to achieve dynamic geological classification. Based on the classification results, the present invention uses an attention - based bidirectional LSTM model to identify the jacking resistance level and optimizes the equipment parameters in combination with historical data. At the same time, the present invention monitors the deviation in real - time and triggers gradient compensation, and matches the best formulation plan according to the mud penetration characteristics. In addition, the present invention also has a self - learning optimization function, which can update the model based on the prediction error, inject prior knowledge through heterogeneous data fusion, and dynamically adjust the decision threshold during the pipe - jacking segmentation. This method significantly improves the intelligent level and accuracy of pipe - jacking construction, effectively copes with complex geological conditions, and improves the construction efficiency and quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flowchart of a micro - pipe - jacking construction regulation method based on multi - source data fusion of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] To further understand the content of the present invention, the present invention will be described in detail in combination with the drawings and embodiments. The following further describes the present application in detail with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than limiting the invention. In addition, it should be noted that, for the sake of description, only the parts related to the invention are shown in the drawings.

[0025] As Figure 1 , a micro - pipe - jacking construction regulation method based on multi - source data fusion in this embodiment may specifically include:

[0026] S101. Obtain the operation parameters of the pipe - jacking equipment, geological parameters, and environmental data through multi - source sensors to generate a multi - source data set, including: obtain the operation parameters of the equipment through an inclination sensor, earth pressure gauge, and torque meter installed on the pipe - jacking equipment; obtain the geotechnical mechanics parameters through a geological detector; obtain the groundwater level and surface vibration data through an environmental monitoring device; obtain the mud viscosity parameters through a rheometer, and generate a formation permeability curve in combination with the data of a permeability tester; integrate the operation parameters of the equipment, the geotechnical mechanics parameters, and the groundwater level and surface vibration data into a multi - source heterogeneous data set.

[0027] Obtain the operation parameters of the equipment through an inclination sensor, earth pressure gauge, and torque meter installed on the pipe - jacking equipment. These sensors can monitor the state of the equipment in real - time. The inclination sensor is used to measure the inclination angle of the equipment to ensure that the pipe - jacking machine remains horizontal or at a preset angle during tunneling, avoiding construction quality problems caused by angle deviation. For example, during pipe - jacking tunneling, when the inclination sensor monitors that the inclination angle of the equipment is 0.5°, exceeding the allowable range of 0.3°, the system will immediately issue an alarm and adjust the equipment posture.

[0028] The earth pressure gauge is used to measure the pressure exerted by the soil on the equipment during tunneling, helping to judge the stability of the strata. For example, when the earth pressure gauge shows a pressure value of 0.8 MPa, it indicates that the soil layer is relatively dense, and the tunneling speed needs to be adjusted to prevent equipment overload. The torque meter is used to monitor the rotational torque of the equipment, reflecting the working load of the equipment. When the torque value reaches the set threshold, the system will automatically reduce the tunneling speed to avoid equipment damage.

[0029] Geotechnical mechanical parameters are obtained through a geological detector. These parameters include the shear strength, compression modulus, and internal friction angle of the soil layer, etc. For example, the geological detector measures the shear strength of the soil layer at a certain depth to be 50 kPa, the compression modulus to be 10 MPa, and the internal friction angle to be 25°. These data are used to evaluate the stability of the strata and the difficulty of tunneling.

[0030] The geological detector detects the strata structure through ultrasonic or electromagnetic wave technology, generates a strata profile diagram, and provides a scientific basis for construction. For example, in a clay layer, the detector shows that the strata structure is uniform, while in a sand layer, there are local loose areas that require reinforcement measures.

[0031] The groundwater level and surface vibration data are obtained through an environmental monitoring device. These data are used to evaluate the impact of construction on the surrounding environment. The groundwater level monitoring device measures the water level depth through a pressure sensor. For example, when the water level is measured to be 1.5 m at a depth of 10 m underground, it indicates that the groundwater in this area is rich, and dewatering measures need to be taken.

[0032] The surface vibration monitoring device measures the vibration intensity through an acceleration sensor. For example, when the vibration acceleration is measured to be 0.1 g on the surface, it indicates that the impact of construction on surrounding buildings is relatively small, but continuous monitoring is required to prevent vibration from intensifying.

[0033] The mud viscosity parameter is obtained through a rheometer, and the strata permeability curve is generated by combining the data of the permeability tester. The rheometer is used to measure the rheological properties of the mud. For example, when the measured mud viscosity is 30 cP, it indicates that the mud has good fluidity and is suitable for use in the sand layer.

[0034] The permeability tester generates a permeability curve by measuring the penetration speed of the mud in the strata. For example, when the measured permeability in the clay layer is 10^-6 m / s, it indicates that the permeability of the mud is poor, and the mud formula needs to be adjusted to improve the penetration effect.

[0035] The equipment operation parameters, geotechnical mechanical parameters, and groundwater level and surface vibration data are integrated into a multi-source heterogeneous data set. These data come from different sources and have different formats, and need to be uniformly processed. For example, the equipment operation parameters are stored in the form of a time series, the geotechnical mechanical parameters are stored in the form of a table, and the groundwater level data are stored in the form of a continuous signal.

[0036] Through data cleaning and format conversion, these data are integrated into a unified dataset, providing a basis for subsequent analysis. The integrated dataset can comprehensively reflect the equipment status, formation characteristics, and environmental impacts during the construction process, providing support for dynamic geological classification and parameter optimization.

[0037] S102. The spatio-temporal alignment and data cleaning of the multi-source dataset are performed to obtain a standardized dataset, including: performing spatio-temporal alignment processing on the multi-source dataset using the distributed Kalman filtering algorithm; removing the abnormal data in the multi-source dataset through the outlier rejection algorithm based on the 3σ principle;

[0038] filling the missing data in the multi-source dataset through the KNN missing value imputation algorithm; generating a standardized dataset with a unified timestamp.

[0039] The distributed Kalman filtering algorithm is used for spatio-temporal alignment processing of the multi-source dataset. Its core idea is to synchronize the data of multiple sensors in time and space to eliminate data inconsistencies caused by different sampling frequencies or data transmission delays. For example, during the pipe jacking construction process, the inclinometer and the earth pressure gauge may collect data at frequencies of once per second and once every two seconds respectively, and there may be a millisecond-level delay in data transmission. Through the distributed Kalman filtering, these data can be unified to a single time reference, ensuring that the data for subsequent analysis has a consistent timestamp. This processing method can effectively avoid analysis errors caused by time misalignment and improve the reliability of the data.

[0040] The outlier rejection algorithm based on the 3σ principle is used to remove the abnormal data in the multi-source dataset. The 3σ principle is a statistical method. Assuming that the data follows a normal distribution, it is considered that the data within the range of the mean ± 3 times the standard deviation is normal, and the data outside this range is an outlier. For example, when collecting the torque data of the pipe jacking equipment, if a certain measurement value is significantly higher than the mean of the historical data and exceeds the 3 times standard deviation range, then this data can be determined as an outlier and removed. This method can effectively filter out the error data caused by sensor failures or external interferences, ensuring the accuracy and consistency of the dataset.

[0041] The KNN missing value imputation algorithm is used to fill in the missing data in multi-source datasets. The KNN algorithm calculates the distance between the missing data points and the neighboring data points, and selects the nearest K data points to perform weighted average or directly take the mean value to fill in the missing values. For example, when collecting groundwater level data, due to sensor failures or data transmission interruptions, the data at some time points may be missing. Through the KNN algorithm, the missing values can be reasonably estimated based on the water level data at neighboring time points, ensuring the integrity of the dataset. This method can reduce the analysis bias caused by data missing and improve the data utilization rate. Generating a standardized dataset with a unified timestamp is an important goal of data processing.

[0042] Through the above steps, the multi-source datasets are aligned in time, the abnormal data is removed, and the missing data is filled, finally generating a standardized dataset. For example, in the monitoring system of pipe jacking construction, data such as inclination angle, soil pressure, and torque are stored with a timestamp once per second after being processed, forming a complete and consistent dataset. This standardized dataset provides a high-quality basis for subsequent feature extraction and analysis, improving the accuracy of geological classification and equipment control.

[0043] S103. Extracting features from the standardized dataset through a neural network model and performing classification to output a dynamic geological classification result, including: inputting the standardized dataset into a convolutional recurrent neural network to extract geological spatial distribution features, equipment power spectrum features, and environmental time-varying features; establishing a feature correlation matrix for the extracted features through the fuzzy C-means clustering algorithm; and outputting a dynamic geological classification result including sand layers, clay layers, and composite strata according to the feature correlation matrix.

[0044] Input the standardized dataset into a convolutional recurrent neural network to extract geological spatial distribution features, equipment power spectrum features, and environmental time-varying features. The convolutional recurrent neural network combines the spatial feature extraction ability of the convolutional neural network and the time series processing ability of the recurrent neural network, and can effectively capture the spatial distribution law of geological data, the power change trend during equipment operation, and the time-varying characteristics of environmental parameters. For example, the geological spatial distribution features can extract the spatial correlation of geotechnical mechanical parameters through the convolutional layer, the equipment power spectrum features can analyze the periodic changes of torque meter and earth pressure gauge data through the recurrent neural network, and the environmental time-varying features can capture the dynamic changes of groundwater level and surface vibration through time series modeling. This multi-dimensional feature extraction method can comprehensively reflect the geological conditions and equipment operation status, providing a reliable basis for subsequent classification.

[0045] The convolutional recurrent neural network consists of an input layer, a convolutional layer, a recurrent layer, and an output layer. In the input layer, the dimension of the standardized data set is the number of time steps × the number of features. The number of features includes geological parameters (such as shear strength, compression modulus), equipment parameters (thrust force, torque), and environmental parameters (groundwater level, vibration acceleration). The convolutional layer uses 3 one-dimensional convolutional kernels (kernel size = 5, stride = 1), the activation function is ReLU, and the dimension of the output feature map is the number of time steps × 32. The recurrent layer contains bidirectional LSTM units (the number of hidden layer nodes = 64), and the attention mechanism focuses on key parameters such as geotechnical shear strength by calculating feature weights. The output layer uses a Softmax classifier to output the probability distributions of sand layers, clay layers, and composite strata. During the training process, the Adam optimizer (learning rate = 0.001) is used, and the loss function is cross-entropy. The training data set contains 100 groups of historical construction data (covering different geological conditions).

[0046] A feature correlation matrix is established for the extracted features through the fuzzy C-means clustering algorithm. Fuzzy C-means clustering is a soft clustering method that allows data points to belong to multiple categories with a certain degree of membership and is suitable for cases with fuzzy boundaries in geological classification. For example, in the extracted geological spatial distribution features, the geotechnical mechanical parameters in some areas may have the characteristics of both sand layers and clay layers at the same time. Fuzzy C-means clustering can calculate the similarity between features, construct a feature correlation matrix, and clarify the degree of association between each feature. This clustering method can effectively handle the uncertainty in geological classification and improve the accuracy of classification results.

[0047] The dynamic geological classification results including sand layers, clay layers, and composite strata are output according to the feature correlation matrix. The feature correlation matrix reflects the correlation between different features. By combining the comprehensive analysis of geological spatial distribution features, equipment power spectrum features, and environmental time-varying features, the dynamic identification of geological categories can be achieved. For example, when the feature correlation matrix shows that the geological spatial distribution features in a certain area are highly correlated with sand layers, and at the same time the equipment power spectrum features and environmental time-varying features also support this conclusion, the system will classify this area as a sand layer. For composite strata, the feature correlation matrix will show that the feature correlations of multiple categories are relatively high, thus outputting the classification results of composite strata. This dynamic classification method can adapt to the changes in complex geological conditions and provide accurate geological information support for pipe jacking construction.

[0048] S104. Identify the jacking resistance level based on the dynamic geological classification results, and determine the equipment control parameters through an optimization algorithm, including: inputting the dynamic geological classification results into a pre-trained working condition identification model, where the working condition identification model is constructed using a bidirectional LSTM network with an attention mechanism; outputting the jacking resistance level label according to the comparison between the geotechnical shear strength characteristic value and the dynamic decision threshold; and based on the jacking resistance level label, calling the historical construction database and using an improved particle swarm algorithm with equipment power constraints and jacking speed constraints to calculate the Pareto optimal solution set of the jacking force and the rotation speed.

[0049] When inputting the dynamic geological classification results into the pre-trained working condition identification model, it is first necessary to clarify the specific content of the dynamic geological classification results. Dynamic geological classification usually includes multi-dimensional information such as geotechnical type, stratum structure, and groundwater status, which are collected and processed in real time through sensors or geological exploration equipment. The pre-trained working condition identification model uses a bidirectional LSTM network with an attention mechanism. This network structure can capture the temporal dependence and spatial correlation in geological data. The introduction of the attention mechanism enables the model to pay more attention to key features, such as the geotechnical shear strength characteristic value, thereby improving the accuracy of identification. For example, when there is a rock stratum with high shear strength in the stratum, the model will focus on analyzing the data in this area through the attention mechanism to ensure that the output jacking resistance level label is more accurate.

[0050] The process of outputting the jacking resistance level label according to the comparison between the geotechnical shear strength characteristic value and the dynamic decision threshold involves the setting of the dynamic decision threshold. The dynamic decision threshold is usually dynamically adjusted based on historical construction data and geological conditions to adapt to different working conditions. For example, when the geotechnical shear strength characteristic value exceeds the threshold, the model will output a higher jacking resistance level label, indicating that the current stratum has a greater resistance to the jacking equipment and corresponding control measures need to be taken. The key to this step lies in the reasonable setting of the dynamic decision threshold to ensure that the model can accurately identify the resistance level under different geological conditions.

[0051] When calling the historical construction database based on the jacking resistance level label, the historical construction database contains records of construction parameters and effects under previous similar geological conditions. By analyzing these data, the optimal control parameters of the equipment under the current resistance level can be determined. For example, when the jacking resistance level is high, the historical data may show that it is necessary to increase the jacking force or reduce the rotation speed to improve construction efficiency. This step makes full use of the experience accumulation of historical data and provides a reliable reference basis for the current construction.

[0052] When using the improved particle swarm optimization algorithm with equipment power constraints and jacking speed constraints to calculate the Pareto optimal solution set of jacking force and rotation speed, the improved particle swarm optimization algorithm ensures that the solved parameters meet the construction requirements and do not exceed the safe operating range of the equipment by introducing the constraint conditions of equipment power and jacking speed. The objective function of the improved particle swarm optimization algorithm is: min F jacking force + ω ⋅ Δτ torque, and the constraint conditions are: P < 50KW, V ≥ 0.2m / min, where ω is the weight coefficient, which is determined by fitting historical data (ω = 0.3).

[0053] The Pareto optimal solution set refers to finding a balance among multiple objectives, such as finding the optimal combination between jacking force and rotation speed to maximize construction efficiency. For example, when the jacking resistance level is relatively high, the algorithm may recommend a larger jacking force and a lower rotation speed to ensure that the equipment can smoothly penetrate the formation while avoiding equipment overload. This step realizes the precise control of equipment parameters through the optimization algorithm, improving the safety and efficiency of construction.

[0054] S105. Adjusting the equipment control parameters according to the real-time monitoring data and matching the mud formula scheme includes: real-time monitoring of the jacking force deviation and the rotation speed volatility; if the ratio of the jacking force deviation to the rotation speed volatility exceeds the set tolerance, calculating the compensation increment through a gradient descent optimizer; retrieving the formula scheme with the highest matching degree from the pre-established mud formula knowledge base according to the compensated jacking force combined with the mud viscosity - formation permeability curve; updating the knowledge base parameters based on the implementation effect of the mud formula.

[0055] Real-time monitoring of the jacking force deviation and the rotation speed volatility is a key link in equipment control. The data of jacking force F and rotation speed ω are collected in real time through sensors, and the deviation ΔF and volatility δ_ω are calculated. For example, when the real-time value of the jacking force F is 1000kN and the set value is 950kN, the deviation ΔF is 50kN; at the same time, the volatility of the rotation speed ω is 5%, that is, δ_ω = 0.05. By real-time monitoring of these data, abnormal situations in equipment operation can be detected in a timely manner, providing a basis for subsequent adjustments.

[0056] If the ratio of the jacking force deviation to the rotation speed volatility exceeds the set tolerance, the gradient descent optimizer is activated to calculate the compensation increment ΔF'. For example, the set tolerance is 1000. When ΔF / δ_ω = 50 / 0.05 = 1000, it just reaches the tolerance upper limit. At this time, the gradient descent optimizer determines the compensation increment ΔF' by calculating the partial derivative ∂E / ∂F of the formation resistance prediction error function E. Assuming that the optimizer calculates the compensation coefficient α = 0.1, then ΔF' = 0.1×∂E / ∂F, thus realizing the precise compensation of the jacking force.

[0057] According to the compensated jacking force F+ΔF', combined with the mud viscosity - formation permeability curve, retrieve the formulation plan with the highest matching degree from the pre-established mud formulation knowledge base. For example, when the compensated jacking force is 1050 kN, combined with the current formation permeability of 0.02 mD and the mud viscosity of 30 mPa·s, through the fuzzy inference rules of the knowledge base, a formulation plan with the bentonite content β = 8% is retrieved. This step ensures a high degree of matching between the mud formulation and formation characteristics, improving the construction efficiency.

[0058] Updating the knowledge base parameters based on the implementation effect of the mud formulation is an important part of the optimization process. For example, during actual construction, if it is found that the penetration effect of the current mud formulation is better than expected, relevant data (such as formation porosity φ = 0.15, bentonite content β = 8%) can be fed back to the knowledge base to update the parameter relationship β = k·ln(1 / φ), where β is the bentonite content in the mud, φ is the formation porosity, and k is a coefficient selected based on construction experience values.

[0059] By continuously optimizing the knowledge base parameters, the matching accuracy of subsequent formulation plans can be improved, providing more reliable guidance for construction. The above steps form a closed-loop control mechanism through real-time monitoring, optimization calculation, formulation matching, and knowledge base update. For example, during construction, if the formation characteristics change, the equipment parameters and mud formulation can be quickly adjusted through real-time monitoring data to ensure the smooth progress of construction. This dynamic adjustment mechanism not only improves the stability and efficiency of equipment operation but also reduces construction risks, providing strong support for construction under complex geological conditions.

[0060] S106. The triggering of model update and optimization of control parameters by prediction error includes: dividing the real-time construction data stream into sample blocks according to a time window; if the prediction error rate of the working condition identification model exceeds the preset model error threshold on three consecutive sample blocks, the elastic weight consolidation algorithm is used to update the network weights; injecting the output of the updated model as prior knowledge into the feature association matrix, and generating an iteratively optimized set of control parameters through heterogeneous data fusion.

[0061] Dividing the real-time construction data stream into sample blocks is to convert the continuous data stream into discrete time periods for analysis. For example, in pipe jacking construction, the data every 10 minutes can be used as a sample block, including parameters such as thrust, torque, and rotational speed. This segmentation method can effectively capture the dynamic changes during the construction process, facilitating subsequent model analysis and error calculation. If too short a time window is used, it may lead to excessive data fluctuations; too long a time window may mask key changes. Therefore, a reasonable selection of the time window length is the key to ensuring the accuracy of model analysis.

[0062] If the prediction error rate of the working condition recognition model exceeds the preset model error threshold on three consecutive sample blocks, for example, the error rate exceeds 5%, the model update mechanism is triggered. Here, the error rate is calculated by comparing the model prediction value with the actual value. The fact that the error rates of three consecutive sample blocks exceed the standard indicates that the model's performance under the current working condition continues to be poor, which may be due to a significant change in the working condition or the inapplicability of the model parameters. At this time, using the Elastic Weight Consolidation algorithm to update the network weights can quickly adapt to the new working condition without destroying the existing knowledge of the model. For example, in pipe jacking construction, when the formation changes from a sand-bearing layer to a clay layer, the prediction error rate of the model may increase significantly. At this time, updating the network weights through the Elastic Weight Consolidation algorithm can enable the model to better adapt to the new formation conditions. Updating the network weights of the working condition recognition model through the Elastic Weight Consolidation algorithm can protect the important weights of previous tasks while learning the current new tasks to overcome the problem of catastrophic forgetting during the update process of the working condition recognition model. In the Elastic Weight Consolidation algorithm, calculate the importance matrix of the old task (L is the loss function, is the network weight); add a regularization term to the new task loss function: , where λ = 0.5.

[0063] Injecting the output of the updated model as prior knowledge into the feature association matrix is to optimize the generation of subsequent control parameters using the latest prediction results of the model. For example, in the updated model, if it is found that the sensitivity of the thrust to formation disturbance has increased significantly, this information can be injected into the feature association matrix as an important basis for the generation of subsequent control parameters.

[0064] The Fuzzy C-Means clustering algorithm calculates the feature association weights through the membership function to generate the feature association matrix , where is the membership degree of feature i to category j.

[0065] By performing feature space mapping on the newly accessed sensor data through the Principal Component Analysis method, the data from different sources can be unified into the same feature space, which is convenient for subsequent analysis. For example, mapping the data of temperature sensors and pressure sensors into the same feature space can better capture the correlation between the two.

[0066] Generating an iteratively optimized set of control parameters is to dynamically adjust the construction parameters by combining the output of the updated model and the result of heterogeneous data fusion. For example, in pipe jacking construction, if the updated model predicts a high formation disturbance coefficient, the propulsion speed and cutter head rotation speed can be adjusted to reduce the disturbance. This iterative optimization method can ensure that the construction parameters are always matched with the current working condition, improving the construction efficiency and safety.

[0067] By dynamically adjusting the decision threshold and tolerance ratio, the performance of the model can be further optimized. For example, when the formation disturbance coefficient is high, the tolerance ratio can be appropriately reduced to improve the model's sensitivity to anomalies.

[0068] S107. Adjusting the formation disturbance threshold and equipment decision threshold according to the segmented evaluation results includes: when the jacking distance reaches the segmented threshold, obtaining the soil stress change amount and weight coefficient of each monitoring point; calculating the formation disturbance coefficient according to the soil stress change amount, weight coefficient and average jacking force; dynamically adjusting the decision threshold and tolerance ratio parameter in the subsequent stage according to the formation disturbance coefficient; the set tolerance is calculated according to the formation disturbance coefficient of the previous construction segment.

[0069] When the jacking distance reaches the segmented threshold, obtain the soil stress change amount and weight coefficient of each monitoring point. During the jacking construction process, the formation will be disturbed to varying degrees, and the soil stress change amount is collected in real time through monitoring points arranged at different positions. For example, when the jacking distance reaches every 100 meters, the system automatically extracts the soil stress change amounts Δσ_A, Δσ_B, Δσ_C of monitoring points A, B, C, and assigns weight coefficients v_A, v_B, v_C according to the importance of each monitoring point. Monitoring point A is located in the key formation area, and the weight coefficient v_A is set to 0.5. Monitoring points B and C are located in the secondary areas respectively, and the weight coefficients v_B and v_C are set to 0.3 and 0.2. In this way, the formation disturbance situation can be comprehensively reflected and accurate data can be provided for subsequent calculations.

[0070] Calculate the formation disturbance coefficient according to the soil stress change amount, weight coefficient and average jacking force. After obtaining the soil stress change amount and weight coefficient of each monitoring point, combine the average jacking force P_avg of the current construction segment to calculate the formation disturbance coefficient K_d. For example, the soil stress change amount Δσ_A of monitoring point A is 50 kPa, those of monitoring points B and C are 30 kPa and 20 kPa respectively, and the average jacking force P_avg is 1000 kN, then K_d = (50×0.5 + 30×0.3 + 20×0.2) / 1000 = 0.036. The formation disturbance coefficient K_d reflects the degree of formation disturbance. The larger the K_d value, the more serious the formation disturbance and the stricter the construction control measures need to be taken.

[0071] Dynamically adjust the decision threshold and tolerance ratio parameters in the subsequent stage according to the formation disturbance coefficient. Based on the calculated formation disturbance coefficient \(K_d\), the system dynamically adjusts the decision threshold \(\epsilon\) and tolerance ratio parameters in the subsequent construction stage. For example, when \(K_d = 0.036\), the system adjusts the decision threshold \(\epsilon\) from the default value of \(0.05\) to \(0.04\), and the tolerance ratio parameter from \(1.2\) to \(1.1\). This adjustment can better adapt to the changes in formation conditions and avoid equipment overload or construction quality decline caused by excessive formation disturbance. By dynamically adjusting the parameters, the construction safety and efficiency can be improved, and the construction risk can be reduced.

[0072] The set tolerance is calculated according to the formation disturbance coefficient of the previous construction section. After the end of each construction section, the system will reset the tolerance according to the formation disturbance coefficient \(K_d\) of the previous construction section. For example, if \(K_d\) of the previous construction section is \(0.036\), the system sets the tolerance to \(\pm5 kPa\). This method of setting tolerance based on the formation disturbance coefficient can more precisely control the construction parameters and avoid construction deviation caused by sudden changes in formation conditions. By dynamically adjusting the tolerance, the construction process can be ensured to be more stable and the construction quality can be improved.

[0073] An embodiment of the present invention also provides a micro - pipe - jacking construction regulation system for implementing any of the methods described in claims 1 - 8, characterized by including:

[0074] A multi - source data acquisition module, used to obtain the operation parameters of the pipe - jacking equipment, geological parameters, and environmental data through multi - source sensors, and generate a multi - source data set;

[0075] A data pre - processing module, used to perform spatio - temporal alignment and data cleaning on the multi - source data set to obtain a standardized data set;

[0076] A dynamic geological classification module, used to extract features from the standardized data set through a neural network model and perform classification, and output a dynamic geological classification result;

[0077] A resistance regulation module, used to identify the jacking resistance level based on the dynamic geological classification result, and determine the equipment regulation parameters through an optimization algorithm;

[0078] A real - time decision - making module, used to adjust the equipment regulation parameters according to real - time monitoring data and match the mud formula scheme;

[0079] A model iteration module, used to trigger model update and optimize the regulation parameters through prediction errors;

[0080] A threshold optimization module, used to adjust the decision threshold according to the segmented evaluation result.

[0081] Although the present invention has been described in detail above with general descriptions and specific embodiments, on the basis of the present invention, some modifications or improvements can be made, which are obvious to those skilled in the art. Therefore, these modifications or improvements made without departing from the spirit of the present invention all fall within the scope of protection required by the present invention.

Claims

1. A micro-pipe jacking construction regulation method based on multi-source data fusion, characterized in that Including: Obtain the operation parameters, geological parameters and environmental data of the pipe jacking equipment through multi-source sensors, and generate a multi-source data set; Perform spatio-temporal alignment and data cleaning on the multi-source data set to obtain a standardized data set; Extract features from the standardized data set through a neural network model and perform classification, and output a dynamic geological classification result; Identify the jacking resistance level based on the dynamic geological classification result, and determine the equipment regulation parameters through an optimization algorithm; Adjust the equipment regulation parameters according to the real-time monitoring data and match the mud formula scheme; Trigger model update and optimize the regulation parameters through the prediction error; Adjust the decision threshold according to the segmented evaluation result.

2. The micro-pipe jacking construction control method according to claim 1, characterized in that, The obtaining of the operation parameters, geological parameters and environmental data of the pipe jacking equipment through multi-source sensors and generating a multi-source data set includes: Obtain the operation parameters of the equipment through the inclinometer, earth pressure gauge and torque meter installed on the pipe jacking equipment; Obtain the geotechnical mechanics parameters through a geological detector; Obtain the groundwater level and surface vibration data through an environmental monitoring device; Obtain the mud viscosity parameters through a rheometer, and generate a formation permeability curve in combination with the data of a permeability tester; Integrate the equipment operation parameters, the geotechnical mechanics parameters and the groundwater level and surface vibration data into a multi-source heterogeneous data set.

3. The micro-pipe jacking construction control method according to claim 1, characterized in that, The performing of spatio-temporal alignment and data cleaning on the multi-source data set to obtain a standardized data set includes: Perform spatio-temporal alignment processing on the multi-source data set by using a distributed Kalman filtering algorithm; Remove the abnormal data in the multi-source data set through an outlier rejection algorithm based on the 3σ principle; Fill in the missing data in the multi-source data set through a KNN missing value imputation algorithm; Generate a standardized data set with a unified timestamp.

4. The micro-pipe jacking construction control method according to claim 3, characterized in that, The extracting of features from the standardized data set through a neural network model and performing classification, and outputting a dynamic geological classification result includes: Input the standardized data set into a convolutional recurrent neural network to extract the geological spatial distribution features, equipment power spectrum features and environmental time-varying features; Establish a feature correlation matrix for the extracted features through a fuzzy C-means clustering algorithm; Output a dynamic geological classification result including sand layer, clay layer and composite formation according to the feature correlation matrix.

5. The micro-pipe jacking construction control method according to claim 1, wherein The identifying of the jacking resistance level based on the dynamic geological classification result and determining the equipment regulation parameters through an optimization algorithm includes: Input the dynamic geological classification result into a pre-trained working condition identification model, and the working condition identification model is constructed by using a bidirectional LSTM network including an attention mechanism; Output a jacking resistance level label according to the comparison between the geotechnical shear strength characteristic value and the dynamic decision threshold; Call the historical construction database based on the jacking resistance level label, and calculate the Pareto optimal solution set of the jacking force and rotation speed by using an improved particle swarm algorithm with equipment power constraint and jacking speed constraint.

6. The micro-pipe jacking construction control method according to claim 1, wherein The adjusting of the equipment regulation parameters according to the real-time monitoring data and matching the mud formula scheme includes: Real-time monitor the jacking force deviation and rotation speed volatility; If the ratio of the jacking force deviation to the rotation speed volatility exceeds the set tolerance, calculate the compensation increment through a gradient descent optimizer; Retrieve the formulation solution with the highest matching degree from the pre-established mud formulation knowledge base according to the compensated jacking force in combination with the mud viscosity - formation permeability curve; Update the knowledge base parameters based on the implementation effect of the mud formulation.

7. The micro-pipe jacking construction control method according to claim 4, wherein, The triggering of model update and optimization of regulation parameters through prediction error includes: Segment the real-time construction data stream into sample blocks according to a time window; If the prediction error rate of the working condition identification model exceeds the preset model error threshold on three consecutive sample blocks, update the network weights using the elastic weight consolidation algorithm; Inject the output of the updated model as prior knowledge into the feature correlation matrix, and generate an iteratively optimized set of regulation parameters through heterogeneous data fusion.

8. The micro-pipe jacking construction control method according to claim 1, wherein, The adjustment of the formation disturbance threshold and equipment decision threshold according to the segmented evaluation results includes: When the jacking distance reaches the segmented threshold, obtain the soil stress change amount and weight coefficient of each monitoring point; Calculate the formation disturbance coefficient according to the soil stress change amount, weight coefficient and average jacking force; Dynamically adjust the decision threshold and tolerance ratio parameter in the subsequent stage according to the formation disturbance coefficient; The set tolerance is calculated according to the formation disturbance coefficient of the previous construction section.

9. A micro-pipe jacking construction control system for implementing any of the methods described in claims 1-8, characterized in that, It includes: A multi-source data acquisition module, which is used to obtain the operation parameters of the pipe jacking equipment, geological parameters and environmental data through multi-source sensors, and generate a multi-source data set; A data preprocessing module, which is used to perform spatio-temporal alignment and data cleaning on the multi-source data set to obtain a standardized data set; A dynamic geological classification module, which is used to extract features from the standardized data set through a neural network model and perform classification, and output a dynamic geological classification result; A resistance regulation module, which is used to identify the jacking resistance level based on the dynamic geological classification result, and determine the equipment regulation parameters through an optimization algorithm; A real-time decision-making module, which is used to adjust the equipment regulation parameters according to real-time monitoring data and match the mud formulation solution; A model iteration module, which is used to trigger model update and optimize regulation parameters through prediction error; A threshold optimization module, which is used to adjust the decision threshold according to the segmented evaluation result.

Images