Island area foundation pit deformation monitoring and predicting system and method

Abstract

The invention belongs to the technical field of ocean engineering new energy, and particularly relates to an island area foundation pit deformation monitoring and predicting system and method.The island area foundation pit deformation monitoring and predicting system comprises a distributed sensor module which is deployed at the corresponding position of a foundation pit and used for monitoring the displacement, the inclination angle, the soil pressure and the underground water level of the foundation pit in real time; the data processing module is used for completing integrated data preprocessing and feature extraction according to embedded sensor groups on the basis of edge computing nodes to obtain foundation pit real-time data; the terminal server receives the real-time data of the foundation pit, a lightweight prediction model is built in the terminal server, and the lightweight prediction model is used for performing multi-source data fusion and collaborative prediction according to the deformation data of the foundation pit; and if the terminal server judges that the foundation pit in the island area deforms, the early warning module gives out early warning. The shallow neural network model is used for predicting whether the foundation pit collapses or not, and the safety of the ocean engineering new energy field in the island area foundation pit construction process is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy for ocean engineering, and particularly relates to a deformation monitoring and prediction system and method for foundation pits in island areas. Background Technique

[0002] Ocean engineering refers to the application of engineering technology to solve various engineering problems related to the marine environment, resources, energy, transportation, etc. It covers multiple fields, including but not limited to offshore oil and gas exploitation, offshore wind power, port construction, offshore platforms, submarine pipelines, and marine environmental protection. Ocean engineering is a national key project. It is not only an important field to promote economic development but also an important part to enhance the country's comprehensive competitiveness and strategic security. Ocean engineering is one of the key fields for national development, and the projects involved cover multiple aspects such as energy development, marine resource utilization, maritime transportation, environmental protection, and national defense security.

[0003] During the construction of ocean engineering, especially during the construction of the foundation pits of infrastructure in new energy fields such as wind power generation, it is necessary to monitor and predict in real time whether the foundation pits in the island area are deformed. Traditional foundation pit monitoring relies on manual regular measurements or real-time data transmission to the cloud for analysis. However, in the island area, there are problems such as poor network signals, unstable power supply, and complex environments, resulting in low real-time data and delayed warnings. Existing prediction models are mostly based on centralized servers, highly dependent on the network, and difficult to adapt to the offline or weak network scenarios in the island area. There is a lack of a fusion analysis and dynamic prediction mechanism for multi-source heterogeneous data (such as geology, hydrology, and structural stress) in the island area. Summary of the Invention

[0004] The purpose of the present invention is to provide a deformation monitoring and prediction system and method for foundation pits in island areas, which can monitor and predict in real time whether the foundation pits in the island area are deformed during the construction process.

[0005] The technical solutions adopted by the present invention are specifically as follows: An island area foundation pit deformation monitoring and prediction system, comprising: A distributed sensor module: Deployed at the corresponding positions of the foundation pit for real-time monitoring of the displacement, inclination angle, soil pressure, and underground water level of the foundation pit; including a displacement meter, an inclinometer, a soil pressure cell, and an underground water level gauge; A data processing module: Based on edge computing nodes, integrated data preprocessing and feature extraction are completed according to the embedded sensor groups to obtain real-time data of the foundation pit; Terminal server: Deployed at the construction site, it receives the real-time foundation pit data output by the edge nodes. The terminal server is built-in with a lightweight prediction model, which is used for multi-source data fusion and collaborative prediction based on the foundation pit deformation data. An adaptive communication module is provided inside the terminal server. The adaptive communication module dynamically switches the transmission mode according to the network status. The transmission modes include: 4G network transmission, 5G network transmission, Wi-Fi transmission, and offline caching. The 4G network transmission, 5G network transmission, and Wi-Fi transmission support resume after network interruption.

[0006] Early warning module: The early warning module is communicatively connected to the terminal server. If the terminal server determines that the foundation pit in the isolated area is deformed, the terminal server sends an early warning control signal to the early warning module. After receiving the early warning control signal, the early warning module issues an early warning; it provides audible and visual alarms and risk level push (such as text messages, local broadcasts).

[0007] Preferably, the distributed sensor module includes: Sensors: including displacement gauges, inclinometers, earth pressure cells, and groundwater level gauges; Data acquisition unit: including a wireless transmission module and an ADC analog-to-digital converter. Each sensor is connected to a wireless transmission module, and the ADC analog-to-digital converter is connected to a wireless reception module and a wireless transmission module; Power supply module: used to supply power to the sensors and the data acquisition unit.

[0008] Preferably, in the data processing module, the process of preprocessing data based on the edge computing node includes: Data cleaning: Remove noise and invalid data to ensure data quality; Data filtering: Screen specific data according to requirements, such as removing outliers; Data compression: Compress the data to be transmitted to reduce the bandwidth consumption of data transmission; Format conversion: Convert the data into string format, and perform normalization processing and standardization processing on the data; Preferably, in the data processing module, a data temporary storage module is built into the edge computer, and the edge computer is temporarily stored in the local storage device, especially when the data needs to be transmitted in batches later, or when further analysis needs to be performed locally. Local storage can reduce the dependence on centralized storage and provide higher data security and response speed.

[0009] Preferably, the lightweight prediction model uses a shallow neural network (Shallow Neural Network, SNN). During the training process of the shallow neural network: First: Prepare the sensor data for training, including the sensor data of normal foundation pits and collapsed foundation pits. Label the data through data tags, then normalize the data, and divide the data into a training set, a validation set, and a test set. The common ratio is 70% for the training set, 15% for the validation set, and 15% for the test set. If there are missing values in the sensor data, supplement them by interpolation or mean filling; according to the characteristics of the sensor data, select the corresponding features for input, for example: displacement gauge data, inclinometer data, pressure data of earth pressure cells, and groundwater level gauge data under time series; the sensor data also includes the geographical distribution of the foundation pit area; Then: Conduct model design, including an input layer, two hidden layers, and an output layer; the number of neurons in the input layer is equal to the number of features; each input feature corresponds to an input neuron, representing each sensor value; each of the hidden layers contains several neurons, and the ReLU (Rectified Linear Unit) activation function is used for non-linear transformation; the output layer uses one neuron and is combined with the Sigmoid activation function to output a value between 0 and 1, representing the probability of collapse; and the binary cross-entropy function is used as the loss function, and the Adam optimizer is used as the model optimizer; Then, conduct model training, configure the training parameters, adjust the batch size, usually set to 32, 64, or 128, depending on the hardware resources and the size of the dataset, set the learning rate, the common initial learning rate is 0.001, which can be dynamically adjusted through a learning rate scheduler, set the number of training epochs, first set to 20 - 50 epochs, and adjust according to the validation error during the training process; add a Dropout layer to the hidden layer to randomly discard a certain proportion of neurons to prevent overfitting; train the model: use the training set to train the model and update the weights through the backpropagation algorithm; validation set evaluation: use the validation set to monitor the performance of the model during the training process to avoid overfitting, and then adjust the hyperparameters: adjust the hyperparameters of the batch size, learning rate, and the number of neurons in the hidden layer according to the loss value and accuracy during the training process; Then, conduct model performance evaluation, evaluate the accuracy and precision of the model through the test set. If the dataset is small, it is recommended to conduct cross-validation; for example, K-fold cross-validation; to improve the stability and generalization ability of the model; reduce the complexity of the model through means such as L2 regularization, and remove unimportant neurons and connections through pruning operations to reduce the complexity of the model; Finally, export and save the trained shallow neural network model.

[0010] Preferably, the terminal server directly calls the trained shallow neural network model to predict the preprocessed real-time foundation pit data. If the collapse probability in the output result exceeds 0.3, a warning control signal is generated. If the collapse probability in the output result does not exceed 0.3, the system continues to monitor and predict.

[0011] Preferably, the warning module further includes a full-process monitoring unit. The full-process monitoring unit is used to monitor whether the data processing module and the terminal server can receive data. If the data processing module or the terminal server cannot receive data, the warning module issues a system operation exception prompt.

[0012] An island area foundation pit deformation monitoring and prediction method; specifically includes the following steps: Step 1: System preparation stage; set up a distributed sensor module; and construct a data processing module, a terminal server, and a warning module; build a lightweight prediction model in the terminal server, complete the training, and output the trained lightweight prediction model to complete the construction of the monitoring and prediction system; Step 2: The distributed sensor module obtains sensing data and sends it to the data processing module through wireless transmission. The data processing module, based on the edge computing node, completes the integrated data preprocessing and feature extraction according to the embedded sensor groups to obtain the real-time foundation pit data, and sends it to the terminal server; Step 3: The terminal server uses the lightweight prediction model to predict the foundation pit collapse probability according to the monitored real-time foundation pit data. If the collapse probability in the output result exceeds 0.3, a warning control signal is generated. If the collapse probability in the output result does not exceed 0.3, the system continues to monitor and predict; Step 4: After receiving the warning control signal, the warning module issues a warning.

[0013] The technical effects achieved by the present invention are: In the present invention, through the edge-local collaborative architecture: solve the problem of unstable network in the island area and reduce the dependence on the cloud; in the present invention, multi-modal data fusion: combine geological parameters with real-time monitoring data to improve the prediction accuracy. In the present invention, through the deployment of a lightweight model: adopt the model quantization technology to enable the shallow neural network model to predict whether the foundation pit collapses, ensuring the safety of the foundation pit construction process in the island area of the marine engineering new energy field. Brief Description of the Drawings

[0014] Figure 1 is the system block diagram of an island area foundation pit deformation monitoring and prediction system of the present invention; Figure 2 is the operation flow chart of an island area foundation pit deformation monitoring and prediction system of the present invention. Detailed Embodiments

[0015] In order to make the objectives and advantages of the present invention more clear and understandable, the present invention will be specifically described below in conjunction with embodiments. It should be understood that the following text is only used to describe one or several specific implementation manners of the present invention, and does not strictly limit the scope of protection specifically claimed by the present invention. Embodiment 1

[0016] As Figure 1 - Figure 2 shown, an island area foundation pit deformation monitoring and prediction system includes: Distributed sensor module: Deployed at the corresponding positions of the foundation pit, used to monitor the displacement, inclination angle, soil pressure and underground water level of the foundation pit in real time; including displacement gauges, inclinometers, soil pressure cells, and underground water level gauges; Data processing module: Based on edge computing nodes, integrated data preprocessing and feature extraction are completed according to the embedded sensor groups to obtain real-time foundation pit data; Terminal server: Deployed at the construction site, receiving the real-time foundation pit data output by the edge node. The terminal server is built-in with a lightweight prediction model, and the lightweight prediction model is used for multi-source data fusion and collaborative prediction according to the foundation pit deformation data; an adaptive communication module is provided inside the terminal server. The adaptive communication module dynamically switches the transmission mode according to the network status. The transmission modes include: 4G network transmission, 5G network transmission, Wi-Fi transmission, and offline caching. The 4G network transmission, 5G network transmission, and Wi-Fi transmission are all interrupted and resumed.

[0017] Early warning module: The early warning module is communicatively connected to the terminal server. If the terminal server determines that the foundation pit in the island area is deformed, the terminal server sends an early warning control signal to the early warning module, and the early warning module issues an early warning after receiving the early warning control signal; providing sound and light alarms, risk level push, such as text messages, local broadcasts.

[0018] In the present invention, through the edge-local collaborative architecture: solve the problem of unstable network in the island area and reduce the dependence on the cloud; in the present invention, multi-modal data fusion: combine geological parameters with real-time monitoring data to improve the prediction accuracy. In the present invention, through the deployment of a lightweight model: adopt model quantization technology to make the shallow neural network model predict whether the foundation pit will collapse, ensuring the safety of the foundation pit construction process in the ocean engineering new energy field in the island area.

[0019] Preferably, the distributed sensor module includes: Sensors: including displacement gauges, inclinometers, soil pressure cells, and underground water level gauges; Data acquisition unit: including a wireless transmission module and an ADC analog-to-digital converter. Each sensor is connected to a wireless transmission module, and the ADC analog-to-digital converter is connected to a wireless receiving module and a wireless transmission module; Power supply module: used to supply power to the sensors and the data acquisition unit.

[0020] Preferably, in the data processing module, the process of preprocessing data based on the edge computing node includes: Data cleaning: removing noise and invalid data to ensure data quality; Data filtering: screening specific data according to requirements, such as removing outliers; Data compression: compressing the data to be transmitted to reduce the bandwidth consumption of data transmission; Format conversion: converting the data into string format, and performing normalization and standardization processing on the data; Preferably, in the data processing module, a data temporary storage module is built in the edge computer, and the edge computer is temporarily stored in the local storage device, especially when the data needs to be transmitted in batches later, or needs to be further analyzed locally. Local storage can reduce the dependence on centralized storage and provide higher data security and response speed.

[0021] Preferably, the lightweight prediction model uses a shallow neural network (SNN). During the training process of the shallow neural network: First: Prepare the sensor data for training, including the sensor data of normal foundation pits and collapsed foundation pits, label the data through data tags, then perform normalization processing on the data, and divide the data into a training set, a validation set, and a test set. The common ratio is 70% for the training set, 15% for the validation set, and 15% for the test set. If there are missing values in the sensor data, supplement them by interpolation or mean filling; according to the characteristics of the sensor data, select the corresponding features for input, for example: displacement gauge data, inclinometer data, pressure data of soil pressure cells, and groundwater level gauge data under time series; the sensor data also includes the geographical distribution of the foundation pit area; Then: conduct model design, including an input layer, two hidden layers, and an output layer; the number of neurons in the input layer is equal to the number of features; each input feature corresponds to an input neuron, representing each sensor value; each of the hidden layers contains several neurons, and the ReLU (Rectified Linear Unit) activation function is used for nonlinear transformation; the output layer uses one neuron and is combined with the Sigmoid activation function to output a value between 0 and 1, representing the probability of collapse; and the binary cross-entropy function is used as the loss function, and the Adam optimizer is used as the model optimizer; Then, perform model training, configure training parameters, adjust the batch size, usually set to 32, 64, or 128, depending on the hardware resources and the size of the dataset, set the learning rate, the common initial learning rate is 0.001, which can be dynamically adjusted through the learning rate scheduler, set the number of training epochs, initially set to 20 - 50 epochs, and adjust according to the validation error during the training process; add a Dropout layer to the hidden layer to randomly discard a certain proportion of neurons to prevent overfitting; train the model: use the training set to train the model and update the weights through the backpropagation algorithm; validation set evaluation: use the validation set to monitor the performance of the model during training to avoid overfitting, and then adjust the hyperparameters: adjust the hyperparameters of the batch size, learning rate, and the number of neurons in the hidden layer according to the loss value and accuracy during the training process; Then, perform model performance evaluation. Evaluate the accuracy and precision of the model through the test set. If the dataset is small, it is recommended to perform cross-validation, such as K-fold cross-validation, to improve the stability and generalization ability of the model; reduce the complexity of the model through means such as L2 regularization, and remove unimportant neurons and connections through pruning operations to reduce the complexity of the model; Finally, export and save the trained shallow neural network model.

[0022] Preferably, the terminal server directly calls the trained shallow neural network model to predict the preprocessed real-time data of the foundation pit. If the collapse probability in the output result exceeds 0.3, a warning control signal is generated. If the collapse probability in the output result does not exceed 0.3, the system continues to monitor and predict.

[0023] Preferably, the warning module further includes a full-process monitoring unit, and the full-process monitoring unit is used to monitor whether the data processing module and the terminal server can receive data. If the data processing module or the terminal server cannot receive data, the warning module issues a system operation exception prompt. Embodiment 2

[0024] As Figure 1 and Figure 2 shown, an island area foundation pit deformation monitoring and prediction method; specifically includes the following steps: Step 1: System preparation stage; set up a distributed sensor module; and construct a data processing module, a terminal server, and a warning module; construct a lightweight prediction model in the terminal server, complete the training, and output the trained lightweight prediction model to complete the construction of the monitoring and prediction system; Step 2: The distributed sensor module acquires sensing data and transmits it to the data processing module via wireless transmission. Based on the edge computing nodes, the data processing module completes the integrated data preprocessing and feature extraction according to each embedded sensor group to obtain the real-time foundation pit data, and then sends it to the terminal server; Step 3: The terminal server uses a lightweight prediction model to predict the foundation pit collapse probability based on the monitored real-time foundation pit data. If the collapse probability in the output result exceeds 0.3, a warning control signal is generated. If the collapse probability in the output result does not exceed 0.3, the system continues to monitor and predict; Step 4: After receiving the warning control signal, the warning module issues a warning.

[0025] In the present invention, through the edge-local collaborative architecture: the problem of unstable network in the isolated area is solved, and the dependence on the cloud is reduced; in the present invention, multi-modal data fusion: combining geological parameters with real-time monitoring data to improve the prediction accuracy. In the present invention, through the deployment of a lightweight model: using model quantization technology, the shallow neural network model is used to predict whether the foundation pit collapses, ensuring the safety of the foundation pit construction in the isolated area in the field of marine engineering new energy.

[0026] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. The structures, devices, and operation methods not specifically described and explained in the present invention, unless otherwise specified and limited, are implemented according to the conventional means in the art.

Claims

1. An island area foundation pit deformation monitoring and prediction system, characterized in that: Including: Distributed sensor module: Deployed at the corresponding positions of the foundation pit for real-time monitoring of the displacement, inclination angle, earth pressure and groundwater level of the foundation pit; Data processing module: Based on the edge computing node, integrated data preprocessing and feature extraction are completed according to each embedded sensor group to obtain real-time foundation pit data; Terminal server: Deployed at the construction site, receiving the real-time foundation pit data output by the edge node. The terminal server is built-in with a lightweight prediction model, and the lightweight prediction model is used for multi-source data fusion and collaborative prediction according to the foundation pit deformation data; An adaptive communication module is provided inside the terminal server. The adaptive communication module dynamically switches the transmission mode according to the network state. The transmission modes include: 4G network transmission, 5G network transmission, Wi-Fi transmission and offline caching. The 4G network transmission, 5G network transmission and Wi-Fi transmission can all resume transmission after disconnection; Early warning module: The early warning module is communicatively connected to the terminal server. If the terminal server determines that the foundation pit in the island area is deformed, the terminal server sends an early warning control signal to the early warning module, and the early warning module issues an early warning after receiving the early warning control signal.

2. The deformation monitoring and prediction system for foundation pits in the island area according to claim 1, characterized in that: The distributed sensor module includes: Sensors: including displacement gauges, inclinometers, earth pressure cells, groundwater level gauges; Data acquisition unit: including a wireless transmission module and an ADC analog-to-digital converter. Each sensor is connected to a wireless transmission module, and the ADC analog-to-digital converter is connected to a wireless reception module and a wireless transmission module; Power module: used to supply power to the sensors and the data acquisition unit.

3. The deformation monitoring and prediction system for foundation pits in the island area according to claim 2, characterized in that: In the data processing module, the process of data preprocessing based on the edge computing node includes: Data cleaning: removing noise and invalid data; Data filtering: screening specific data according to requirements; Data compression: compressing the data to be transmitted; Format conversion: converting the data into a string format, and performing normalization processing and standardization processing on the data.

4. An island area foundation pit deformation monitoring and prediction system according to claim 3, characterized in that: In the data processing module, a data temporary storage module is built in the edge computer, and the edge computer is temporarily stored in the local storage device.

5. The deformation monitoring and prediction system for foundation pits in an island area according to claim 4, wherein: The lightweight prediction model uses a shallow neural network. During the training of the shallow neural network: First: Prepare the sensor data for training, including the sensor data of normal foundation pits and collapsed foundation pits, label them through data tags, then perform normalization processing on the data, and divide the data into training set, validation set and test set; According to the characteristics of the sensor data, select the corresponding features for input; Then: Carry out model design, including an input layer, two hidden layers and an output layer; The number of neurons in the input layer is equal to the number of features; Each input feature corresponds to an input neuron, representing each sensor value; Each of the hidden layers contains several neurons, and the ReLU activation function is used for non-linear transformation; The output layer uses a single neuron and is combined with the Sigmoid activation function to output a value between 0 and 1, representing the probability of collapse; And use the binary cross-entropy function as the loss function and the Adam optimizer as the model optimizer; Then, perform model training, configure training parameters, adjust the batch size, set the learning rate, and set the number of training epochs; Train the model: Use the training set to train the model and update the weights through the backpropagation algorithm. Validation set evaluation: Use the validation set to monitor the performance of the model during training to avoid overfitting, and then adjust the hyperparameters: According to the loss value and accuracy during training, adjust the hyperparameters of the batch size, learning rate, and the number of neurons in the hidden layer. Then, perform model performance evaluation, and evaluate the accuracy and precision of the model through the test set. Finally, export and save the trained shallow neural network model.

6. The deformation monitoring and prediction system for foundation pits in an island area according to claim 5, characterized in that: The terminal server directly calls the trained shallow neural network model to predict the preprocessed real-time data of the foundation pit. If the collapse probability in the output result exceeds 0.3, a warning control signal is generated. If the collapse probability in the output result does not exceed 0.3, the system continues to monitor and predict.

7. An island area foundation pit deformation monitoring and prediction system according to claim 6, characterized in that: The warning module further includes a full-process monitoring unit, which is used to monitor whether the data processing module and the terminal server can receive data. If the data processing module or the terminal server cannot receive data, the warning module issues a system operation exception prompt.

8. A method for predicting the deformation monitoring of a foundation pit in an isolated island area, characterized in that: The monitoring and prediction method is the operation method of an island area foundation pit deformation monitoring and prediction system according to any one of claims 1-7; specifically includes the following steps: Step 1: System preparation stage; Set up a distributed sensor module; and construct a data processing module, a terminal server, and a warning module; Build a lightweight prediction model in the terminal server, complete the training, and output the trained lightweight prediction model to complete the construction of the monitoring and prediction system. Step 2: The distributed sensor module obtains sensing data and sends it to the data processing module through wireless transmission. The data processing module, based on the edge computing node, completes the integrated data preprocessing and feature extraction according to each embedded sensor group to obtain the real-time data of the foundation pit, and sends it to the terminal server. Step 3: The terminal server uses the lightweight prediction model to predict the foundation pit collapse probability according to the monitored real-time data of the foundation pit. If the collapse probability in the output result exceeds 0.3, a warning control signal is generated. If the collapse probability in the output result does not exceed 0.3, the system continues to monitor and predict. Step 4: After receiving the warning control signal, the warning module issues a warning.

Images