Construction method for drilling holes for bridge pile foundations in karst areas

Abstract

The present application provides a construction method for drilling holes for bridge pile foundations in karst areas, and relates to a construction method technology for drilling holes for bridge pile foundations. The method comprises the following steps: S1, collecting historical data of geological surveys of pile foundation locations during historical construction, wherein the historical data include geological feature images and geological data, annotating the geological feature images and geological data, and constructing a cave feature prediction model based on deep learning; S2, optimizing the construction strategy based on reinforcement learning, defining the state and action space of the reinforcement learning model, designing a reward function to comprehensively consider the factors of construction, and verifying the optimized construction strategy in a simulation environment by training a deep Q network; S3, deploying sensors at the construction site to collect real-time data of the real-time status, and inputting the reinforcement learning model after fusion with the historical data, and outputting the optimal construction strategy based on the real-time data to ensure efficient and safe construction.

Description

Technical Field

[0001] The invention relates to the technical field of bridge pile foundation hole forming construction methods, in particular to a bridge pile foundation hole forming construction method used in karst areas. Background Art

[0002] During the construction of bridge pile foundations, karst areas are often encountered. Karst areas refer to special landforms formed by the long-term dissolution of soluble rocks such as limestone and dolomite by corrosive water. In karst areas, the structure of the rock mass changes due to karst action, which leads to reduced rock strength and increased permeability, which usually easily causes problems such as insufficient bearing capacity of the foundation, collapse of the roof of the underlying cave, ground subsidence, and uneven settlement of the foundation. The characteristics of caves in karst areas are diverse, and caves with different characteristics have different impacts on pile foundation construction. However, unified treatment measures are often used for caves with different characteristics, resulting in the risk of collapse of some caves.

[0003] The patent with the prior art publication number CN115613556A provides a construction method for drilling bridge pile foundations in karst areas. The method conducts detailed geological exploration of the pile foundation location in the karst area, records the specific conditions of the geology and karst caves, and then classifies and takes corresponding treatment measures according to the different characteristics of the karst caves. Before construction, the outer casing is used to pre-protect the strong weathering surface, and then the punching operation is performed to protect the soil around the pile hole from being affected, increase the friction between the soil outside the pile hole and the outer casing, thereby improving the bearing capacity of the pile foundation poured later. The center of the tube mouth of the outer casing needs to be aligned with the center of the pile foundation, and is fixedly connected to the steel plate of the opening casing to enhance the stability of the outer casing. A grouting hole is drilled in the soil around the karst cave pile, and cement slurry is injected through the grouting pipe to improve the connectivity and friction between the outer casing and the soil, and further strengthen the bearing capacity of the pile foundation. The bottom end of the outer casing is also equipped with a soil extrusion drill bit, which helps to form a compaction layer close to the outer casing and enhance the friction between the outer casing and the soil.

[0004] Although this method has solved some of the construction problems caused by caves to a certain extent, it still has some limitations. This method mainly relies on the results of geological exploration and manual judgment to formulate construction strategies. For complex and changeable karst areas, this construction method based on experience and manual judgment is difficult to ensure the safety and efficiency of construction. In addition, this method lacks effective response strategies for real-time geological changes and dynamic adjustments during the construction process, which may lead to increased risks and costs during the construction process.

[0005] In view of this, we proposed a construction method for drilling holes for bridge pile foundations in karst areas. Summary of the invention

[0006] The invention provides a construction method for drilling holes for bridge pile foundations in karst areas, and the technical solution includes the following aspects.

[0007] According to one aspect of an embodiment of the present application, a construction method for drilling holes for bridge pile foundations in karst areas is provided, and the method includes:

[0008] S1. Collect historical data of geological surveys of pile foundation locations during historical construction, wherein the historical data includes geological feature images and geological data, annotate the geological feature images and geological data, and construct a karst cave feature prediction model based on deep learning.

[0009] The specific steps of constructing the cave feature prediction model are as follows: the input layer receives geological feature images and geological data.

[0010] The geological feature image is input into the convolutional neural network, and the convolution layer extracts spatial features through multi-layer convolution operations.

[0011] h i，j l =σ(∑ m，n w m，n l x i+m，j+n l-1 +b l ), where h i，j l is the output of the lth convolutional layer, w m，n l is the convolution kernel, x i+m，j+n l-1 is the input of the l-1th layer, b l is the bias and σ is the activation function.

[0012] The pooling layer reduces the size of the geological feature image through pooling operations, extracts the main features, and inputs the spatial features and main features into the fully connected layer to form a high-dimensional feature representation.

[0013] The geological data is input into the fully connected layer to form high-dimensional feature representation 2, and the self-attention mechanism is introduced to fuse the high-dimensional feature representation 1 and the high-dimensional feature representation 2 to form a fused feature.

[0014] The output layer outputs predicted attribute features, and the output layer includes a classification prediction output layer and a regression prediction output layer.

[0015] The classification prediction output layer classifies and predicts the shape and overlying layer type of the cave, and the regression prediction output layer performs regression prediction on the depth, height, and area of ​​the cave.

[0016] The classification prediction output layer uses the cross entropy loss function as the loss function of the model, and the regression prediction output layer loss function is the smooth L1 loss function. The classification loss and regression loss are combined to optimize the feature extraction capability of the network structure.

[0017] Use the training set to train the model, the validation set to evaluate the model performance, input new geological feature images and geological data, and use the trained model to make predictions.

[0018] S2. Optimize the construction strategy based on reinforcement learning, define the state and action space of the reinforcement learning model, design the reward function to comprehensively consider the construction factors, and verify the optimized construction strategy in a simulation environment by training the deep Q network.

[0019] S3. Sensors are deployed at the construction site to collect real-time data of the real-time status, which is then integrated with historical data and input into the reinforcement learning model. The optimal construction strategy is output based on the real-time data to ensure efficient and safe construction.

[0020] S4. Summarize and evaluate after construction, record construction data in detail, compare actual and expected results to evaluate construction decisions, evaluate construction quality and efficiency through pile foundation quality inspection, and continue to optimize models and strategies.

[0021] Among them, in step S1, geological survey data of the pile foundation position during historical construction is collected, and the geological feature images include geological radar images and seismic exploration images; the geological data include electrical logging data and sonic logging data.

[0022] The contents of the annotation of the geological feature images and geological data are the attribute characteristics of the cave, and the attribute characteristics include shape, overburden type, depth, height, and area.

[0023] Among them, the self-attention mechanism is introduced to calculate the self-attention weights of the attribute features of geological feature images and geological data respectively; image (Q image , K image , V image )=softmax(Q image K image T / d k 1 / 2 )V image , Attention data (Q data , K data , V data )=softmax(Q data K data T / d k 1 / 2 )V data , where Q, K, and V represent query, key, and value, respectively. k Represents the dimension of the key, and the softmax function is used to normalize the attention weights.

[0024] Perform feature fusion to obtain historical data feature FImage , F Image =∑ i Attention image (Q image , K image , V image )f i , F Data =∑ j Attention data (Q data , K data , V data )f j ; The fusion representation of the geological feature image and the geological data is spliced ​​to form the fusion feature F 融合 , F 融合 =Concat(F Image , F Data ).

[0025] Among them, the reinforcement learning model includes state space and action space.

[0026] The state space of the reinforcement learning model and the extracted fusion feature F 融合 and real-time data as the state input of reinforcement learning, the state is represented by s t =[F 融合 , v t ,s twall ] Among them, v t is the mud loss rate, s twall It is the stability state of the hole wall.

[0027] Design the action space of the reinforcement learning model, action a t For construction strategy.

[0028] Among them, the reward function involves construction efficiency, cost, safety, quality of pile foundation and mud loss control.

[0029] The reward function is expressed as, R=w 1 ×Efficiency+w 2 ×(-Cost)+w 3 ×Safety+w 4 ×Quality+w 5 ×(-Leakage).

[0030] Among them, w iis a weight factor, indicating the importance of each goal; Efficiency is a construction efficiency reward, including positive rewards based on the number of pile foundations completed per unit time or the percentage of construction progress; Cost is a cost control reward, which penalizes decisions that lead to cost increases, including excessive use of materials, excessive rework or equipment failure; Safety is a safety reward, including hole wall collapse and mud leakage; Quality is a pile foundation quality reward, including pile foundation bearing capacity, hole wall integrity and pile foundation verticality; Leakage is a mud loss control reward, taking measures to reduce mud loss.

[0031] Among them, the optimal construction strategy is output according to the real-time status; sensors are deployed at the construction site to collect real-time data of the real-time status.

[0032] The real-time data is transmitted to the central control system, and the real-time data is pre-processed by the geological data acquisition module to obtain the real-time mud loss velocity v t and real-time hole wall stability state s twall .

[0033] Obtain real-time geological radar images and seismic exploration images based on geological anomaly monitoring equipment.

[0034] The real-time image fusion feature F is extracted through the convolutional neural network part of the cave feature prediction model. Image (t), and the historical data feature F Data Splice to get the real-time fusion feature F 融合 (t)=Concat(F Image (t), F Data ).

[0035] The current state s t =[F 融合 (t), v t ,s twall ] is input into the reinforcement learning model, and the reinforcement learning model is based on the current state s t Output the Q value of each action and select the action a with the largest Q value t As the current optimal construction strategy.

[0036] According to the optimal action a output by the reinforcement learning model t , adjust the construction strategy in real time; among them, monitor the construction effect after adjustment in real time, and collect data such as the new round of mud loss rate, drilling depth and hole wall stability.

[0037] Feed the new data into the reinforcement learning model to update the current state s t+1 , and execute the real-time monitoring and adjustment process cyclically.

[0038] Among them, summary and evaluation: record and store all key data during the construction process, organize the collected data by chronological order and category, generate a detailed record of the construction process, including the status, actions and results at each time point, compare the actual results with the expected results at each time point, and evaluate the effectiveness of each construction decision.

[0039] Calculate construction efficiency and total cost, evaluate the economy and efficiency of the entire construction process, and continue to use real-time monitoring and dynamic adjustment mechanisms in future construction projects.

[0040] Collect new construction data, continuously iterate and optimize models and strategies, continue to review and optimize after construction is completed, feed back the experience of each construction into the model, form a virtuous circle, and gradually improve the level of intelligent construction.

[0041] A construction system for drilling holes for bridge pile foundations in karst areas, the system comprising:

[0042] The geological data acquisition module is used to acquire and organize data related to the geological characteristics of karst areas, including geological exploration reports, historical construction data, geological radar images and seismic exploration images, to obtain geological characteristic images and geological data, and provide input for the cave characteristic prediction model.

[0043] The cave feature prediction model module builds a cave feature prediction model based on a deep learning model, which is used to predict geological anomalies in karst areas. The model receives data provided by the geological data acquisition module and uses convolutional neural networks and other related algorithms to extract and fuse features of geological radar images and seismic exploration images, thereby identifying the potential attribute characteristics of the cave.

[0044] The central control system receives real-time data from sensors and monitoring equipment, as well as prediction results from the cave characteristic prediction model. The central control system uses a reinforcement learning model to output the optimal construction strategy based on the current geological and construction conditions. The central control system is also responsible for real-time monitoring of key indicators during the construction process, and adjusting construction strategies based on real-time data to ensure the safety and efficiency of the construction process.

[0045] The construction strategy execution module adjusts the construction operation in real time by controlling the corresponding construction equipment according to the optimal construction strategy output by the central control system.

[0046] The system also includes a user interaction interface for displaying real-time data of the construction process, prediction results, optimal construction strategies and other information, which is convenient for construction managers to monitor and make decisions. The user interaction interface also provides data export and report generation functions, which is convenient for construction managers to review and optimize the construction process.

[0047] The present application provides a construction method for drilling holes for bridge pile foundations in karst areas, which has the following advantages.

[0048] By using the geological prediction model trained by historical geological data and geological characteristics of karst areas, the geological conditions of the pile foundation location can be accurately predicted, providing a scientific basis for construction decisions and reducing construction risks. The application of reinforcement learning algorithms enables the system to intelligently recommend the optimal construction strategy according to different geological conditions and refine the construction parameters, thereby improving construction efficiency and safety. Real-time collection of on-site data, dynamic adjustment of construction strategies through feedback mechanisms, ensure the smooth progress of the construction process, and respond to geological anomalies and potential problems in a timely manner to avoid accidents. After the construction is completed, the system automatically reviews the entire construction process, analyzes the effectiveness of construction decisions, and improves the intelligence level of future construction through algorithm iterative optimization, achieving continuous improvement of construction technology. The design of the user interface enables engineers and technicians to intuitively view geological prediction results, recommended construction strategies, real-time monitoring data and construction reports, facilitating fine-tuning and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flow chart of a construction method for drilling holes for bridge pile foundations in karst areas is provided as an embodiment of the present application.

[0050] Figure 2 A structural diagram of a construction system for drilling holes for bridge pile foundations in karst areas provided in one embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0052] Please refer to Figure 1 , which shows a flow chart of a construction method for drilling holes for bridge pile foundations in karst areas provided by an embodiment of the present application.

[0053] S1. Collect historical data of geological surveys of pile foundation locations during historical construction, wherein the historical data includes geological feature images and geological data, annotate the geological feature images and geological data, and construct a karst cave feature prediction model based on deep learning.

[0054] The specific steps of constructing the cave feature prediction model are as follows: the input layer receives geological feature images and geological data.

[0055] The geological feature image is input into the convolutional neural network, and the convolution layer extracts spatial features through multi-layer convolution operations: h i，j l =σ(∑ m，n w m，n l xi+m，j+n l-1 +b l ), where h i，j l is the output of the lth convolutional layer, w m，n l is the convolution kernel, x i+m，j+n l-1 is the input of the l-1th layer, b l is the bias and σ is the activation function.

[0056] The pooling layer reduces the size of the geological feature image through pooling operations, extracts the main features, and inputs the spatial features and main features into the fully connected layer to form a high-dimensional feature representation.

[0057] The geological data is input into the fully connected layer to form high-dimensional feature representation 2, and the self-attention mechanism is introduced to fuse the high-dimensional feature representation 1 and the high-dimensional feature representation 2 to form a fused feature.

[0058] The output layer outputs predicted attribute features, and the output layer includes a classification prediction output layer and a regression prediction output layer.

[0059] The classification prediction output layer classifies and predicts the shape and overlying layer type of the cave, and the regression prediction output layer performs regression prediction on the depth, height, and area of ​​the cave.

[0060] The classification prediction output layer uses the cross entropy loss function as the loss function of the model, and the regression prediction output layer loss function is the smooth L1 loss function. The classification loss and regression loss are combined to optimize the feature extraction capability of the network structure.

[0061] Use the training set to train the model, the validation set to evaluate the model performance, input new geological feature images and geological data, and use the trained model to make predictions.

[0062] S2. Optimize the construction strategy based on reinforcement learning, define the state and action space of the reinforcement learning model, design the reward function to comprehensively consider the construction factors, and verify the optimized construction strategy in a simulation environment by training the deep Q network.

[0063] S3. Sensors are deployed at the construction site to collect real-time data of the real-time status, which is then integrated with historical data and input into the reinforcement learning model. The optimal construction strategy is output based on the real-time data to ensure efficient and safe construction.

[0064] S4. Summarize and evaluate after construction, record construction data in detail, compare actual and expected results to evaluate construction decisions, evaluate construction quality and efficiency through pile foundation quality inspection, and continue to optimize models and strategies.

[0065] As an optional solution of the present application, in step S1, geological survey data of the pile foundation position during historical construction is collected, and the geological feature images include geological radar images and seismic exploration images; the geological data include electrical logging data and sonic logging data.

[0066] The contents of the annotation of the geological feature images and geological data are the attribute characteristics of the cave, and the attribute characteristics include shape, overburden type, depth, height, and area.

[0067] The specific steps of constructing the cave feature prediction model are as follows: the input layer receives geological feature images and geological data.

[0068] The geological feature image is input into the convolutional neural network, and the convolution layer extracts spatial features through multi-layer convolution operations: h i，j l =σ(∑ m，n w m，n l x i+m，j+n l-1 +b l ), where h i，j l is the output of the lth convolutional layer, w m，n l is the convolution kernel, x i+m，j+n l-1 is the input of the l-1th layer, b l is the bias and σ is the activation function.

[0069] The pooling layer reduces the size of the geological feature image through pooling operations, extracts the main features, and inputs the spatial features and main features into the fully connected layer to form a high-dimensional feature representation.

[0070] Among them, the self-attention mechanism is introduced to calculate the self-attention weights of the attribute features of geological feature images and geological data respectively; image (Q image , K image , V image )=softmax(Q image K image T / d k 1 / 2 )V image , Attention data (Q data , K data , V data )=softmax(Q data K data T / d k 1 / 2 )V data, where Q, K, and V represent query, key, and value, respectively. k Represents the dimension of the key, and the softmax function is used to normalize the attention weights.

[0071] Perform feature fusion to obtain historical data feature F Image , F Image =∑ i Attention image (Q image , K image , V image )f i , F Data =∑ j Attention data (Q data , K data , V data )f j ; The fusion representation of the geological feature image and the geological data is spliced ​​to form the fusion feature F 融合 , F 融合 =Concat(F Image , F Data ).

[0072] The output layer outputs predicted attribute features, and the output layer includes a classification prediction output layer and a regression prediction output layer.

[0073] The classification prediction output layer classifies and predicts the shape and overlying layer type of the cave, and the regression prediction output layer regresses and predicts the depth, height, and area of ​​the cave. The activation function of the classification prediction output layer is the softmax activation function, which maps any real number to the interval (0, 1) and the sum is 1. The softmax function formula is y k =exp(a k ) / ∑ n i=1 exp(a i ), a k is the kth input signal in the output layer, y k is the kth output signal, and n is the number of neurons in the output layer. The cross entropy loss function is used as the loss function of the model. The cross entropy loss function formula is L classification =-∑ n k=1 t k logy k , where t k is the true attribute feature of the kth category, y kis the predicted probability of the kth category (obtained by the softmax function), n is the number of categories, and the output value is converted into a probability distribution to indicate the possibility of each category.

[0074] The activation function of the regression prediction output layer is the linear function f(x)=wx+b, and the loss function is the smooth L1 loss function L regression =1 / n∑|y-y'|, where y is the true value, y' is the predicted value, n is the number of samples, comprehensive classification loss (cross entropy) and regression loss (mean square error), optimize the feature extraction ability of the network structure, L total =αL classification +βL regression , use the training set to train the model, the validation set to evaluate the model performance, input new geological feature images and exploration data, and use the trained model to make predictions.

[0075] As an optional solution of the present application, in step S2, the state space of the reinforcement learning model is defined, the state space of the reinforcement learning model, the extracted fusion feature F 融合 and real-time data as the state input of reinforcement learning, the state is represented by s t =[F 融合 , v t ,s twall ] Among them, v t is the mud loss rate, s twall It is the stability state of the hole wall.

[0076] Design the action space of the reinforcement learning model, action a t It is the construction strategy; such as the selection of cave treatment measures, such as full filling, semi-filling or no filling, whether to use inner and outer casings, and the adjustment of construction parameters; the construction parameters are the burial depth of the outer casing, the size of the punching stroke, the mud density control, etc.

[0077] The reward function involves construction efficiency, cost, safety, pile foundation quality and mud loss control; the reward function is expressed as, R = w 1 ×Efficiency+w 2 ×(-Cost)+w 3 ×Safety+w 4 ×Quality+w 5 ×(-Leakage).

[0078] Among them, w iis a weight factor, indicating the importance of each goal; Efficiency is a construction efficiency reward, including positive rewards based on the number of pile foundations completed per unit time or the percentage of construction progress; Cost is a cost control reward, which penalizes decisions that lead to cost increases, including excessive use of materials, excessive rework or equipment failure; Safety is a safety reward, including hole wall collapse and mud leakage; Quality is a pile foundation quality reward, including pile foundation bearing capacity, hole wall integrity and pile foundation verticality.

[0079] Leakage is a reward for mud loss control, and measures are taken to reduce mud loss; the central control system is encouraged to take measures to reduce mud loss, such as adjusting the mud density and adding materials in a timely manner; a deep Q network is built to map the state to the Q value of the action, and the input layer inputs the state s t The hidden layer includes multiple fully connected layers, using the ReLU activation function, and the output layer outputs the Q value of each action. The loss function is L(θ)=E (st，at，rt，st+1)∼D [(r t +γmax a′ Q(s t+1 , a′; θ - )-Q(s t , a t ;θ)) 2 ], where θ is the parameter of the current Q network, θ - is the parameter of the target Q network, γ is the discount factor, during training, initialize the parameters of the current Q network and the target Q network, initialize the experience replay pool D, sample the state, action, reward and next state in the environment, store them in the experience replay pool D, and randomly sample a small batch (s t , a t , r t ,s t+1 ), calculate the target Q value, y t =r t +γmaxa′Q(s t+1 , a′; θ - ), use the mean square error to update the parameters of the current Q network, L(θ)=E[(y t -Q(s t , a t ;θ)) 2 ], every several steps, the parameters of the current Q network are copied to the target Q network, and the construction process under different geological conditions is simulated through the simulation platform to verify the performance of the optimized reinforcement learning model and construction strategy in various complex situations.

[0080] As an optional solution of the present application, in step S3, the optimal construction strategy is output according to the real-time status.

[0081] Sensors are deployed at the construction site to collect real-time data of the real-time status; the real-time data is transmitted to the central control system, and the real-time data is pre-processed by the geological data acquisition module to obtain the real-time mud loss rate v t and real-time hole wall stability state s twall .

[0082] Obtain real-time geological radar images and seismic exploration images through geological anomaly monitoring equipment. Extract real-time image fusion features F through the convolutional neural network part of the cave feature prediction model Image (t), and the historical data feature F Data Splice to get the real-time fusion feature F 融合 (t)=Concat(F Image (t), F Data ).

[0083] The current state s t =[F 融合 (t), v t ,s twall ] is input into the reinforcement learning model, and the reinforcement learning model is based on the current state s t Output the Q value of each action and select the action a with the largest Q value t As the current optimal construction strategy.

[0084] According to the optimal action a output by the reinforcement learning model t , adjust construction strategies in real time.

[0085] Construction strategies such as adjusting the mud density, increasing or decreasing the water content or additives in the mud by controlling the mud pump, changing the mud density to control the mud loss rate, increasing the backfill volume, judging whether it is necessary to increase the backfill volume of stone and clay according to the real-time monitoring data to stabilize the hole wall and reduce mud leakage, adjusting the punching stroke, changing the stroke size of the punching equipment to reduce the impact on the hole wall and avoid hole wall collapse; real-time monitoring of the adjusted construction effect, collecting a new round of mud loss rate, drilling depth, hole wall stability and other data, feeding the new data back to the reinforcement learning model, and updating the current state s t+1 , and execute the real-time monitoring and adjustment process cyclically.

[0086] Deploy a variety of sensors at the construction site; mud loss rate sensors monitor mud loss rate; borehole depth sensors record changes in borehole depth in real time; geological anomaly monitoring equipment, such as seismic wave sensors and geological radars, monitor geological anomalies, such as the risk of cave roof collapse; use wireless networks or wired connections to transmit sensor data to the central control system in real time, and the data is pre-processed by the data acquisition system, such as noise filtering and data format conversion, to obtain real-time mud loss rate v based on data from different sensors.t and real-time hole wall stability state s twall .

[0087] As an optional solution of the present application, in step S4; Summarize and evaluate: record and store all key data during the construction process (such as mud loss rate, drilling depth, geological anomalies, construction strategies used, etc.); organize the collected data by chronological order and category to generate a detailed record of the construction process, including the status, actions and results at each time point, compare the actual results at each time point with the expected results, and evaluate the effectiveness of each construction decision.

[0088] For example, compare the effects of different mud gravity adjustment strategies on mud loss rate, analyze whether the expected control effect has been achieved, and evaluate key indicators such as the final pile foundation's bearing capacity, stability, and verticality through pile foundation quality inspection after construction.

[0089] Calculate construction efficiency and total cost, evaluate the economy and efficiency of the entire construction process, and continue to use real-time monitoring and dynamic adjustment mechanisms in future construction projects.

[0090] Collect new construction data, continuously iterate and optimize models and strategies, continue to review and optimize after construction is completed, feed back the experience of each construction into the model, form a virtuous circle, and gradually improve the level of intelligent construction.

[0091] Please refer to Figure 2 , which shows a structural diagram of a construction system for drilling holes for bridge pile foundations in karst areas provided by an embodiment of the present application.

[0092] The geological data acquisition module is used to acquire and organize data related to the geological characteristics of karst areas, including geological exploration reports, historical construction data, geological radar images and seismic exploration images, to obtain geological characteristic images and geological data, and provide input for the cave characteristic prediction model.

[0093] The cave feature prediction model module builds a cave feature prediction model based on a deep learning model, which is used to predict geological anomalies in karst areas. The model receives data provided by the geological data acquisition module and uses convolutional neural networks and other related algorithms to extract and fuse features of geological radar images and seismic exploration images, thereby identifying the potential attribute characteristics of the cave.

[0094] The central control system receives real-time data from sensors and monitoring equipment, as well as prediction results from the cave characteristic prediction model. The central control system uses a reinforcement learning model to output the optimal construction strategy based on the current geological and construction conditions. The central control system is also responsible for real-time monitoring of key indicators during the construction process, and adjusting construction strategies based on real-time data to ensure the safety and efficiency of the construction process.

[0095] The construction strategy execution module adjusts the construction operation in real time by controlling the corresponding construction equipment, such as mud pumps, backfill machinery, punching equipment, etc., according to the optimal construction strategy output by the central control system.

[0096] In addition, the system also includes a user interaction interface for displaying real-time data of the construction process, prediction results, optimal construction strategies and other information, which is convenient for construction managers to monitor and make decisions. The user interaction interface also provides data export and report generation functions, which is convenient for construction managers to review and optimize the construction process.

[0097] Through the collaborative work of the above modules, the construction system provided in this application can realize the intelligent construction of the bridge pile foundation drilling process in karst areas, improve construction efficiency, reduce costs and ensure construction safety.

Claims

1. A construction method for drilling holes for bridge pile foundations in karst areas, characterized in that: The method comprises: S1. Collect historical data of geological surveys of pile foundation locations during historical construction, wherein the historical data includes geological feature images and geological data, annotate the geological feature images and geological data, and construct a cave feature prediction model based on deep learning. The specific construction of the cave characteristic prediction model is as follows: The input layer receives geological feature images and geological data; The geological feature image is input into the convolutional neural network, and the convolution layer extracts spatial features through multi-layer convolution operations; h i,j l =σ(∑ m,n w m,n l x i+m,j+n l-1 +b l ), Among them, h i,j l is the output of the lth convolutional layer, w m,n l is the convolution kernel, x i+m,j+n l-1 is the input of the l-1th layer, b l is the bias, σ is the activation function; The pooling layer reduces the size of the geological feature image through pooling operations, extracts the main features, and inputs the spatial features and main features into the fully connected layer to form a high-dimensional feature representation. The geological data is input into the fully connected layer to form high-dimensional feature representation 2, and the self-attention mechanism is introduced to fuse the high-dimensional feature representation 1 and the high-dimensional feature representation 2 to form a fused feature; The output layer outputs predicted attribute features, and the output layer includes a classification prediction output layer and a regression prediction output layer. The classification prediction output layer classifies and predicts the shape and overlying layer type of the cave, and the regression prediction output layer performs regression prediction on the depth, height, and area of ​​the cave; The classification prediction output layer uses the cross entropy loss function as the loss function of the model, and the regression prediction output layer loss function is the smooth L1 loss function. The classification loss and regression loss are combined to optimize the feature extraction ability of the network structure. S2. Optimize construction strategies based on reinforcement learning, define the state space and action space of the reinforcement learning model, design a reward function to comprehensively consider construction factors, and verify the optimized construction strategy in a simulation environment by training a deep Q network; the reinforcement learning model includes a state space and an action space; The state space of the reinforcement learning model and the extracted fusion feature F 融合 and real-time data as the state input of reinforcement learning, the state is represented by s t =[F 融合 ,v t ,s twall ] Among them, v t is the mud loss rate, s twall is the stability state of the hole wall; Design the action space of the reinforcement learning model, action a t For construction strategies; S3. Deploy sensors at the construction site to collect real-time data of the real-time status, integrate it with historical data and input it into the reinforcement learning model. Output the optimal construction strategy based on the real-time data to ensure efficient and safe construction; Deploy sensors at construction sites to collect real-time data on current conditions; The real-time data is transmitted to the central control system, and the real-time data is pre-processed by the geological data acquisition module to obtain the real-time mud loss velocity v t and real-time hole wall stability state s twall , Obtain real-time geological radar images and seismic exploration images based on geological anomaly monitoring equipment; The real-time image fusion feature F is extracted through the convolutional neural network part of the cave feature prediction model. Image (t), and the historical data feature F Data Splice to get the real-time fusion feature F 融合 (t)=Concat(F Image (t),F Data ); The current state s t =[F 融合 (t),v t ,s twall ] is input into the reinforcement learning model, and the reinforcement learning model is based on the current state s t Output the Q value of each action and select the action a with the largest Q value t As the current optimal construction strategy; According to the optimal action at output by the reinforcement learning model, the construction strategy is adjusted in real time.

2. A construction method for drilling holes for bridge pile foundations in karst areas according to claim 1, characterized in that: The method further comprises: S4. Summarize and evaluate after construction, record construction data in detail, compare actual and expected results to evaluate construction decisions, evaluate construction quality and efficiency through pile foundation quality inspection, and continue to optimize models and strategies.

3. A construction method for drilling holes for bridge pile foundations in karst areas according to claim 1, characterized in that: The geological feature images include geological radar images and seismic exploration images; the geological data include electrical logging data and sonic logging data; The contents of the annotation of the geological feature images and geological data are the attribute characteristics of the cave, and the attribute characteristics include shape, overburden type, depth, height, and area.

4. A construction method for drilling holes for bridge pile foundations in karst areas according to claim 2, characterized in that: The self-attention mechanism is introduced to calculate the self-attention weights for the attribute features of geological feature images and geological data respectively; Attention image (Q image ,K image ,V image )=softmax(Q image K image T / d k 1 / 2 )V image , Attention data (Q data ,K data ,V data )=softmax(Q data K data T / d k 1 / 2 )V data , where Q, K, and V represent query, key, and value respectively. k Represents the dimension of the key, and the softmax function is used to normalize the attention weights; Perform feature fusion to obtain historical data feature F Image , F Image =∑ i Attention image (Q image ,K image ,V image )f i , F Data =∑ j Attention data (Q data ,K data ,V data )f j ; The fusion representation of geological feature image and geological data is spliced ​​to form fusion feature F 融合 , F 融合 =Concat(F Image ,F Data ).

5. The construction method for drilling holes for bridge pile foundations in karst areas according to claim 1, characterized in that: The reward function involves construction efficiency, cost, safety, quality of pile foundation and mud loss control; The reward function is expressed as, R = w1×Efficiency+w2×(-Cost)+w3×Safety+w4×Quality+w5×(-Leakage); Among them, w i is the weight factor, indicating the importance of each goal; Efficiency is a reward for construction efficiency, including positive rewards based on the number of pile foundations completed per unit time or the percentage of construction progress; Cost is a reward for cost control, which penalizes decisions that lead to cost increases, including overuse of materials, excessive rework, or equipment failure; Safety is a safety reward, including hole wall collapse and mud leakage; Quality is the reward for pile foundation quality, including pile foundation bearing capacity, hole wall integrity and pile foundation verticality; Leakage is a reward for controlling mud loss and taking measures to reduce mud loss.

6. A construction method for drilling holes for bridge pile foundations in karst areas according to claim 1, characterized in that: Monitor the construction effect after adjustment in real time and collect the new round of mud loss rate, drilling depth and hole wall stability; Feed the new data into the reinforcement learning model to update the current state s t+1 , and execute the real-time monitoring and adjustment process cyclically.

7. A construction method for drilling holes for bridge pile foundations in karst areas according to claim 2, characterized in that: Summary and evaluation: Record and store all key data during the construction process, organize the collected data by time sequence and category, generate detailed records of the construction process, including the status, actions and results at each time point, compare the actual results with the expected results at each time point, and evaluate the effectiveness of each construction decision; Calculate construction efficiency and total cost, evaluate the economy and efficiency of the entire construction process, and continue to use real-time monitoring and dynamic adjustment mechanisms in future construction projects; Collect new construction data, continuously iterate and optimize models and strategies, continue to review and optimize after construction is completed, feed back the experience of each construction into the model, form a virtuous circle, and gradually improve the level of intelligent construction.

8. A construction system for drilling holes for bridge pile foundations in karst areas, applied to the construction method for drilling holes for bridge pile foundations in karst areas as claimed in claim 1, the system comprising: The geological data acquisition module is used to obtain and organize data related to the geological characteristics of karst areas, including geological exploration reports, historical construction data, geological radar images and seismic exploration images, to obtain geological characteristic images and geological data, and provide input for the cave characteristic prediction model; The cave feature prediction model module builds a cave feature prediction model based on a deep learning model to predict geological anomalies in karst areas. The model extracts and fuses features of geological radar images and seismic exploration images by receiving data provided by the geological data acquisition module, thereby identifying the potential attribute characteristics of the cave; The central control system receives real-time data from sensors and monitoring equipment, and the prediction results from the cave feature prediction model. The central control system outputs the optimal construction strategy based on the current geological and construction status through the reinforcement learning model. The central control system is also responsible for real-time monitoring of various key indicators during the construction process, and adjusting the construction strategy based on real-time data to ensure the safety and efficiency of the construction process; The construction strategy execution module adjusts the construction operation in real time by controlling the corresponding construction equipment according to the optimal construction strategy output by the central control system.

9. A construction system for drilling holes for bridge pile foundations in karst areas according to claim 8, characterized in that: The system also includes a user interaction interface for displaying real-time data, prediction results and optimal construction strategies of the construction process, which facilitates construction managers to monitor and make decisions. The user interaction interface also provides data export and report generation functions, which facilitates construction managers to review and optimize the construction process.

Images