Automatic monitoring system for deep foundation pit
Through the deep foundation pit automated monitoring system, using sensors and machine learning technology, real-time monitoring and early warning, the problems of low efficiency and slow response in the existing technology are solved, and construction safety and data accuracy are improved.
Patent Information
- Application Number
- CN202411064250.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-19
AI Technical Summary
The existing deep foundation pit monitoring system relies on manual and traditional measurement methods, has low efficiency, discontinuous data collection, is difficult to capture subtle changes in real time, and is slow to respond, so it is impossible to promptly warn and take countermeasures, resulting in insufficient construction safety and environmental monitoring.
Design a deep foundation pit automated monitoring system, including the perception end, the data processing end and the prediction end, collect parameters in real time through multiple sensors, and use machine learning to build a deep foundation pit state prediction model, real-time monitoring and early warning, and provide it with a graphical interface. Operation function.
Real-time and continuous monitoring of deep foundation pits and surrounding environments is achieved, and potential risks are warned in advance, construction safety and monitoring data are improved, and the reliability of foundation pit projects is ensured.
Smart Images

Figure CN118979526A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of foundation pit monitoring, and in particular to an automatic monitoring system for a deep foundation pit. Background Art
[0002] With the acceleration of urbanization, deep foundation pit projects have played an increasingly important role in construction projects. However, due to its complexity and high risk, deep foundation pit construction poses a major challenge to the surrounding environment and construction safety. In order to ensure the smooth progress of deep foundation pit projects and the safety of the surrounding environment, it is particularly important to monitor the key parameters of deep foundation pits in real time, such as displacement, settlement, stress, groundwater level and soil pressure.
[0003] Existing deep foundation pit monitoring systems mostly rely on manual monitoring and traditional measurement methods, which have many defects. First, manual monitoring is inefficient, data collection is discontinuous, it is difficult to capture subtle changes in the state of the foundation pit in a timely manner, and it is easily affected by human factors, resulting in inaccurate monitoring results. Secondly, traditional measurement methods often cannot reflect the dynamic changes of the state of the foundation pit in real time, and the response speed to emergencies is slow, making it difficult to issue early warnings and take countermeasures in a timely manner. Therefore, we propose an automated monitoring system for deep foundation pits. Summary of the invention
[0004] The purpose of the present invention is to provide an automatic monitoring system for deep foundation pits to solve the problems raised in the above-mentioned background technology.
[0005] To achieve the above-mentioned object, the present invention provides the following technical solutions: an automatic monitoring system for a deep foundation pit, comprising a sensing end, a data processing end, a prediction end and a monitoring end;
[0006] The sensing end includes a variety of sensor components, including an inclinometer component, a settlement meter component, a pressure sensor component, a groundwater level meter component, and an earth pressure meter component; the sensing end is used to collect various parameters of the deep foundation pit and the surrounding environment in real time, including displacement, settlement, stress, groundwater level, and earth pressure, and transmit the collected data to the data processing end;
[0007] The data processing end includes a data collection component and a data processing component; the data collection component is used to receive the original data transmitted by the sensing end, and the data processing component is used to perform data cleaning, denoising and solving on the received data;
[0008] The prediction end includes a model building component, a feature extraction component, a model evaluation component and a model optimization component; the model building component is used to use machine learning technology to build a deep foundation pit state prediction model, which is based on historical monitoring data and deep foundation pit engineering data to predict the state changes of various monitoring parameters of the foundation pit in the future; the feature extraction component is used to select features that have an impact on the foundation pit state prediction from historical monitoring data and deep foundation pit engineering data as model input parameters; the model evaluation component is used to evaluate the performance of the trained deep foundation pit state prediction model; the model optimization component is used to regularly update and optimize the deep foundation pit state prediction model according to new monitoring data or model evaluation results to adapt to changes in the foundation pit construction process;
[0009] The monitoring end includes a real-time monitoring component, a model calling component, an alarm judgment component and a user interaction component; the real-time monitoring component is used to draw the monitoring data transmitted by the data processing end into a real-time curve; the model calling component is used to call the deep foundation pit state prediction model of the prediction end, input the engineering data of the current deep foundation pit, and the model outputs the state changes of various monitoring parameters of the deep foundation pit in a period of time in the future. The model calling component further draws the results of the model output into a prediction curve; the alarm judgment component is used to compare the difference between the real-time curve and the prediction curve in real time, and combine the preset alarm threshold to determine whether the deep foundation pit is in a safe state. If the threshold is exceeded, an alarm message is generated; the user interaction component is used to provide a graphical interface to display the real-time monitoring curve output by the real-time monitoring component, the prediction curve output by the model calling component, and the alarm information generated by the alarm judgment component, and provide interface operation functions to the user.
[0010] Preferably, the data cleaning of the data processing component includes processing of missing values and outliers. For missing values, a linear interpolation algorithm is used to fill them; for outliers, a threshold judgment method is used to regard data points that exceed a preset threshold range as outliers and replace them with the mean of their neighboring points.
[0011] The data processing component uses a sliding average filtering algorithm to perform data denoising, and the algorithm formula is:
[0012]
[0013] Among them, y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window.
[0014] Preferably, the data processing component uses the least square method to perform data calculation to fit the relationship between the sensor data and the actual physical quantity, and its algorithm formula is:
[0015] θ=(X T X) -1 XT Y
[0016] Among them, θ is the fitting parameter, X is the input data matrix, Y is the output data vector, X T is the transposed matrix of X, (X T X) -1 For X T The inverse matrix of X.
[0017] Preferably, the feature extraction component evaluates the importance of the feature by calculating the correlation between each feature and the target variable; the linear correlation between each feature and the target variable is measured using the Pearson correlation coefficient, and the calculation formula of the Pearson correlation coefficient is:
[0018]
[0019] Among them, r represents the correlation coefficient between the feature and the target variable, n represents the number of samples, x represents the feature value, and y represents the target variable value;
[0020] According to the calculated correlation coefficient, the features whose absolute values are greater than the set threshold are selected as useful features to form a feature data set for the subsequent training of the deep foundation pit state prediction model.
[0021] Preferably, the method for the model building component to train the deep foundation pit state prediction model is:
[0022] S1: Collect historical data, which includes the status data of the deep foundation pit project at each monitoring node, including displacement, settlement, stress, groundwater level and soil pressure, and basic data corresponding to the deep foundation pit project, including the project's geographical location, geological conditions, design parameters and construction methods;
[0023] S2: Standardize the cleaned feature data and filter out the features used for model training through the feature extraction component to form a training data set;
[0024] S3: Divide the training data set into a training set and a validation set;
[0025] S4: Select the long short-term memory network LSTM model to build a deep foundation pit state prediction model, and initialize the parameters of the model; the initialized parameters include: initializing the structural parameters of the LSTM model, including the feature dimension of the input layer, the number of hidden layers and the number of hidden units in each layer, and the number of nodes in the output layer, where the feature dimension of the input layer matches the number of extracted deep foundation pit state-related features, and the number of nodes in the output layer is equal to the predicted target number, that is, the state prediction value under all monitoring nodes; initialize the model weights and biases, and use small random numbers for initialization; set the activation function of the LSTM model, including setting the Sigmoid function for the gating mechanism, and setting the Tanh function for the calculation of candidate memory units and hidden states; configure the optimization algorithm, and set the learning rate parameters;
[0026] S5. Model training and evaluation: Use the training set data to iteratively train the selected model. During the training process, the validation set is regularly used to evaluate the model performance until the model performance meets the requirements. The specific method is: Use the training set data to iteratively train the LSTM model. The training set data includes the actual status data of the deep foundation pit project at each monitoring node and the basic data of the deep foundation pit project. In each iteration, the following operations are performed: a) Send a batch of input data to the LSTM model, and update the memory unit and hidden state by calculating the values of the input gate, forget gate and output gate, and finally obtain the predicted output sequence, which is the state change curve of the deep foundation pit project at all monitoring nodes; b) Calculate the loss function value based on the predicted output and the true label; c) Calculate the gradient of the loss function with respect to the model parameters through the back propagation algorithm; d) Use the optimization algorithm to update the model weights and biases according to the gradient;
[0027] S6: After training is completed, the weights and biases of the model are saved, and the model is deployed for subsequent deep foundation pit status prediction.
[0028] Preferably, during the model training process, the mean square error MSE algorithm is used to evaluate the performance of the deep foundation pit state prediction model. The specific algorithm is:
[0029]
[0030] Where n is the number of data points, Y pred,i is the temperature value predicted by the model, Y actual,i is the actual measured temperature value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.
[0031] Preferably, the graphical interface of the user interaction component includes a real-time monitoring curve display area, a prediction curve display area, an alarm information display area, and a user operation function area;
[0032] In the real-time monitoring curve display area, the monitoring data curves of various monitoring parameters of the deep foundation pit transmitted by the real-time monitoring component are displayed in real time, including displacement, settlement, stress, groundwater level and soil pressure;
[0033] In the prediction curve display area, the prediction curve of the state change of each monitoring parameter of the deep foundation pit output by the model calling component in the future period is displayed, which is used for users to compare the current state with the predicted state and identify potential risks in advance;
[0034] In the alarm information display area, the alarm information generated by the alarm judgment component is displayed in real time, including the monitoring parameters that exceed the preset threshold, the alarm level and the recommended measures;
[0035] In the user operation function area, interface operation functions are provided, including zooming and dragging the curve chart to view detailed data, selecting a specific time period for data analysis, and setting alarm thresholds.
[0036] Preferably, before the deep foundation pit state prediction model is deployed, a test data set that is completely different from the training set and the validation set is constructed, and the deep foundation pit state prediction model is tested using the test data set, and the final performance of the model is evaluated by calculating the prediction accuracy, recall rate, and F1 score.
[0037] Compared with the prior art, the present invention has the following beneficial effects:
[0038] The system integrates multiple sensor components through the perception end to achieve real-time and continuous collection of deep foundation pit and surrounding environmental parameters, avoiding the inefficiency and discontinuity of manual monitoring. The prediction end uses machine learning technology to build a deep foundation pit state prediction model, which can predict the state changes of various monitoring parameters of the foundation pit in the future based on historical data, and realize early warning of potential risks. The alarm judgment component compares the difference between the real-time curve and the predicted curve in real time, and generates alarm information in time in combination with the preset alarm threshold, which helps to quickly take countermeasures and ensure construction safety. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 It is the overall structure diagram of the present invention;
[0040] Figure 2 Diagram of the construction and training steps of the deep foundation pit state prediction model;
[0041] Figure 3 Design interface diagrams for user interaction components; DETAILED DESCRIPTION
[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0043] See also Figure 1-3 The present invention provides a technical solution: an automatic monitoring system for a deep foundation pit, including a sensing end, a data processing end, a prediction end and a monitoring end.
[0044] The sensing end is the data acquisition part of the system, which includes a variety of sensor components, specifically inclinometer components, settlement meter components, pressure sensor components, groundwater level meter components and soil pressure meter components. The sensor components can collect various key parameters of the deep foundation pit and the surrounding environment in real time, including displacement, settlement, stress, groundwater level and soil pressure, and then transmit the collected data to the data processing end by wireless or wired means.
[0045] The data processing end is responsible for receiving the raw data transmitted by the sensing end and further processing and analyzing it. The data processing end includes a data collection component and a data processing component. The data collection component is responsible for receiving and storing the raw data from the sensing end to ensure the integrity and continuity of the data. The data processing component cleans, denoises and solves the received data to improve the accuracy and reliability of the data.
[0046] The prediction end is the core part of the system, responsible for building and optimizing the deep foundation pit state prediction model. The prediction end includes model building components, feature extraction components, model evaluation components, and model optimization components. The model building component uses machine learning technology to build a deep foundation pit state prediction model based on historical monitoring data and deep foundation pit engineering data. The model can predict the state changes of various monitoring parameters of the foundation pit in the future. The feature extraction component selects features that have an impact on the foundation pit state prediction from historical monitoring data and deep foundation pit engineering data as model input parameters to improve the prediction accuracy of the model. The model evaluation component is used to evaluate the performance of the trained deep foundation pit state prediction model to ensure the reliability and effectiveness of the model. The model optimization component regularly updates and optimizes the deep foundation pit state prediction model based on new monitoring data or model evaluation results to adapt to changes in the foundation pit construction process.
[0047] The monitoring end is the user interaction part of the system, which is responsible for displaying real-time monitoring data, prediction results and alarm information, and providing interface operation functions. The monitoring end includes real-time monitoring components, model calling components, alarm judgment components and user interaction components. The real-time monitoring component draws the monitoring data transmitted by the data processing end into a real-time curve to intuitively display the dynamic changes of the foundation pit state. The model calling component calls the deep foundation pit state prediction model of the prediction end, inputs the engineering data of the current deep foundation pit, and the model outputs the state changes of each monitoring parameter of the deep foundation pit in the future period of time, and draws the results into a prediction curve. The alarm judgment component compares the difference between the real-time curve and the prediction curve in real time, and combines the preset alarm threshold to determine whether the deep foundation pit is in a safe state. If the threshold is exceeded, an alarm message is generated to remind the user to take corresponding measures. The user interaction component provides a graphical interface to display the real-time monitoring curve, prediction curve and alarm information, and provides users with interface operation functions such as data query and model parameter setting.
[0048] The deep foundation pit automatic monitoring system proposed in the present invention realizes comprehensive and real-time monitoring and early warning of the deep foundation pit and the surrounding environment by integrating the sensing end, data processing end, prediction end and monitoring end, thereby improving the safety and reliability of the deep foundation pit project.
[0049] The present invention will be further described below in conjunction with Examples 1 to 3:
[0050] Embodiment 1:
[0051] The sensor components are used to collect various key parameters of the deep foundation pit and surrounding environment in real time and accurately, providing basic data for subsequent data processing, prediction and monitoring.
[0052] Inclinometer component: Inclinometer is used to measure the inclination of foundation pit support structure, which is one of the important indicators for evaluating foundation pit stability. This system uses high-precision and high-sensitivity inclinometers to ensure that it can capture the slight inclination changes of the support structure.
[0053] Settlement meter components: Settlement meters are used to monitor the settlement of foundation pits and surrounding ground surfaces. This system selects settlement meters that can work stably for a long time to ensure the continuity and reliability of measurement data.
[0054] Pressure sensor assembly: The pressure sensor is used to monitor the stress of the foundation pit support structure. This system selects pressure sensors that can withstand a large pressure range and have high precision to ensure that the stress state of the support structure can be accurately reflected. The pressure sensor should have strong anti-interference ability and good stability to ensure the accuracy of the measurement data.
[0055] Groundwater level meter components: Groundwater level meters are used to monitor the changes in groundwater levels in and around foundation pits. This system uses high-precision, easy-to-install groundwater level meters to ensure that groundwater level data can be obtained in real time.
[0056] Soil pressure gauge assembly: The soil pressure gauge is used to monitor the pressure of the soil around the foundation pit on the supporting structure. This system selects soil pressure gauges that can accurately measure soil pressure to ensure that it can reflect the actual effect of the soil on the supporting structure.
[0057] According to the actual situation and monitoring requirements of deep foundation pit projects, corresponding sensor components are installed at key locations such as foundation pit support structures, surrounding surface and groundwater levels. Sensor installation should ensure stability and reliability to avoid sensor damage or measurement data distortion due to construction disturbances or environmental factors.
[0058] The sensor component transmits the collected data to the data processing end through wired or wireless means. For wired transmission, a dedicated cable needs to be laid to connect the sensor to the data processing end; for wireless transmission, wireless communication technology (such as Zigbee, LoRa, NB-IoT, etc.) is used to achieve remote data transmission. After the sensor is installed, data calibration and verification are required. The sensor is calibrated using a standard signal source or a known measurement value to ensure the accuracy of the measurement data; at the same time, the reliability and consistency of the sensor component can be evaluated by comparing the measurement results of different sensor components or verifying with other monitoring methods (such as manual measurement).
[0059] The data collection component is responsible for receiving and storing the raw data from the sensing end to ensure the integrity and continuity of the data. During the data storage process, the data collection component will timestamp the data for subsequent data processing and analysis.
[0060] The data processing component cleans, denoises and solves the received data to improve the accuracy and reliability of the data. The specific processing steps are as follows:
[0061] The main purpose of data cleaning is to deal with missing values and outliers in the data.
[0062] For missing values, linear interpolation algorithm is used to fill them. The specific steps are: first determine the two valid data points before and after the missing value, then calculate the estimated value of the missing value through linear interpolation based on the values of these two data points, and use the estimated value to fill the missing value.
[0063] For outliers, the threshold judgment method is used for processing. The specific steps are: first, according to historical data and engineering experience, set the reasonable threshold range of each monitoring parameter; then traverse the original data and regard the data points that exceed the preset threshold range as outliers; finally, replace the outlier with the mean of the outlier's neighboring points to eliminate its impact on subsequent analysis.
[0064] In the process of deep foundation pit monitoring, the collected raw data often contains noise due to various reasons such as environmental factors, equipment accuracy, transmission interference, etc. The presence of noise will affect the accuracy of the data and the effectiveness of subsequent analysis. Therefore, it is necessary to denoise the raw data to extract useful signals and improve the credibility of the data.
[0065] This implementation adopts a sliding average filtering algorithm to perform data denoising. The sliding average filtering is a simple and effective denoising method, which smoothes the data by calculating the average value of the data in a certain window, thereby eliminating the interference of random noise. The specific steps of the sliding average filtering algorithm are:
[0066] Step 1. Determine the sliding window size N: The sliding window size N is an important parameter that affects the denoising effect. The selection of N should be determined according to the characteristics of the specific monitoring data and the noise level. The larger the N, the more obvious the denoising effect, but it may cause the loss of data details; the smaller the N, the noise may not be completely eliminated. Therefore, in practical applications, it is necessary to determine the appropriate N value through experiments or experience. When monitoring the flow of groundwater in deep foundation pits, since the flow data may be affected by many factors and fluctuate greatly, N = 12 can be selected as the sliding window size to effectively smooth the data and reduce noise. For monitoring the pressure on the supporting structure, since the pressure data is relatively stable and the fluctuation is small, N = 4 can be selected as the sliding window size, which can smooth the data without losing too much details. When monitoring the liquid level changes in the foundation pit and the surrounding surface, according to the actual situation of the liquid level fluctuation and the noise level, N = 8 can be selected as the sliding window size. This is a moderate choice that can strike a balance between smoothing data and retaining details.
[0067] Step 2, read the original data: read the original data sequence x[n] of deep foundation pit monitoring from the data acquisition system or storage medium, where n represents the index of the data point.
[0068] Step 3: Apply the sliding average filter algorithm:
[0069] For each data point in the original data sequence x[n], the sliding average filter algorithm is applied to calculate its denoised value y[n]. The specific calculation formula is as follows:
[0070]
[0071] Where y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window. The sliding average filtering algorithm calculates the average value of the original data in the sliding window as the denoised data value.
[0072] Step 4: Output the denoised data: Output the calculated denoised data sequence y[n] to a storage medium.
[0073] De-noising the deep foundation pit monitoring data through the sliding average filtering algorithm can effectively eliminate the interference of random noise and improve the accuracy and reliability of the data.
[0074] In the deep foundation pit automatic monitoring system, the data processing component is responsible for converting the raw data collected by the sensor into actual physical quantities for subsequent analysis and use. In order to achieve this conversion, the present invention uses the least squares method to solve the data to fit the relationship between the sensor data and the actual physical quantity. The least squares method is a mathematical optimization technique that finds the best function match for the data by minimizing the sum of squares of errors. In the present invention, we use the least squares method to fit the relationship between the sensor data (input data matrix X) and the actual physical quantity (output data vector Y) to obtain the fitting parameter θ. The algorithm formula of the least squares method is:
[0075] θ=(X T X) -1 X T Y
[0076] Among them, θ is the fitting parameter, X is the input data matrix, Y is the output data vector, X T is the transposed matrix of X, (X T X) -1 For X T The inverse matrix of X. Data solution steps:
[0077] Step 1, data preparation: Get the raw data from the sensor components and organize them into input data matrix X and output data vector Y. Each row of the input data matrix X represents an observation sample, and each column represents a sensor data feature; each element of the output data vector Y represents the actual physical quantity corresponding to the row of the input data matrix X.
[0078] Step 2: Calculate the transposed matrix and product matrix: Calculate the transposed matrix X of the input data matrix X T , calculate the product matrix X T X.
[0079] Step 3: Calculate the inverse matrix: multiply the matrix X T X is inverted and its inverse matrix (X T X) -1 .
[0080] Step 4: Calculate the fitting parameters: Calculate the fitting parameter vector θ according to the least squares algorithm formula. Each element of θ represents the strength and direction of the relationship between the corresponding column in the input data matrix X and the actual physical quantity.
[0081] Step 5, verification and evaluation: Use the fitting parameters θ and the new observation samples of the input data matrix X to make predictions, and compare the prediction results with the actual physical quantities. The accuracy and reliability of the fitting model are evaluated by calculating indicators such as prediction error and residual sum of squares.
[0082] Embodiment 2:
[0083] In the deep foundation pit automatic monitoring system, the feature extraction component is responsible for extracting features with strong correlation with the target variable (such as the stability state of the deep foundation pit) from the original sensor data to form a feature data set for use in the subsequent deep foundation pit state prediction model. The present invention uses the Pearson correlation coefficient to measure the linear correlation between each feature and the target variable, and selects useful features based on the strength of the correlation. The Pearson correlation coefficient is a statistical indicator that measures the degree of linear correlation between two variables, and its value range is [-1,1], where 1 represents a complete positive correlation, -1 represents a complete negative correlation, and 0 represents no correlation. The calculation formula of the Pearson correlation coefficient is:
[0084]
[0085] Among them, r represents the correlation coefficient between the feature and the target variable, n represents the number of samples, x represents the feature value, and y represents the target variable value.
[0086] For each feature, use the Pearson correlation coefficient calculation formula to calculate the correlation coefficient between it and the target variable. Repeat this step until the correlation coefficients between all features and the target variable are calculated.
[0087] According to the calculated correlation coefficient, the features whose absolute value is greater than the set threshold are selected as useful features. The threshold can be adjusted according to the actual application scenario and needs to balance the quantity and quality of the features. For the settlement monitoring of buildings around the foundation pit, the threshold is set to ±10mm or less, depending on factors such as the structural type of the building, the importance level, and the distance between the foundation pit and the building. For the horizontal displacement monitoring of supporting piles or underground continuous walls, the threshold may be set from a few millimeters to tens of millimeters, depending on factors such as the design requirements of the supporting structure, the depth of the foundation pit, and the surrounding geological conditions. Changes in the groundwater level have an important impact on the stability of the foundation pit. Therefore, it is necessary to set a reasonable groundwater level monitoring threshold based on the actual project situation. For example, a threshold for the rate of water level rise or fall can be set, such as not exceeding a certain value per day.
[0088] Finally, the selected useful features are combined into a feature dataset.
[0089] The LSTM model is designed for processing and predicting time series data. In the prediction of deep foundation pit status, the status data (such as displacement, settlement, etc.) changes over time. The LSTM model can effectively capture the long-term dependencies and patterns in these time series data, thereby more accurately predicting the future status of the deep foundation pit. The training method of the deep foundation pit status prediction model based on the LSTM network includes:
[0090] S1: Collect historical data of deep foundation pit projects, including status data at each monitoring node (such as displacement, settlement, stress, groundwater level and soil pressure) and basic data (such as project geographical location, geological conditions, design parameters and construction methods).
[0091] S2: Clean the collected data, remove outliers and missing values, and perform standardization to keep the numerical ranges of different features consistent. Subsequently, the features closely related to the deep foundation pit status are screened out through the feature extraction component to form a training data set.
[0092] S3: Divide the processed training data set into a training set and a validation set. The training set is used for iterative training of the model, and the validation set is used to evaluate the performance of the model to ensure that the model has good generalization ability.
[0093] S4: Select the LSTM model as the deep foundation pit state prediction model and initialize the model parameters. The initialized parameters include the structural parameters of the LSTM model (such as the feature dimension of the input layer, the number of hidden layers and the number of hidden units in each layer, and the number of nodes in the output layer), where the feature dimension of the input layer matches the number of extracted deep foundation pit state-related features, and the number of nodes in the output layer is equal to the predicted target number (i.e., the state prediction value under all monitoring nodes). At the same time, initialize the model weights and biases, and use small random numbers for initialization to avoid the model falling into a local optimal solution at the beginning of training. Set the activation function of the LSTM model, including the Sigmoid function for the gating mechanism and the Tanh function for the calculation of candidate memory units and hidden states. Finally, configure the optimization algorithm (such as the Adam algorithm) and set the appropriate learning rate parameters to control the training speed and stability of the model.
[0094] S5: Iteratively train the selected LSTM model using the training set data. In each iteration, the following operations are performed: a) A batch of input data is fed into the LSTM model, and the memory cells and hidden states are updated by calculating the values of the input gate, forget gate, and output gate, and finally the predicted output sequence (i.e., the state change curve of the deep foundation pit project at all monitoring nodes) is obtained; b) The loss function value is calculated based on the predicted output and the true label to evaluate the prediction error of the model; c) The gradient of the loss function with respect to the model parameters is calculated through the back propagation algorithm; d) The optimization algorithm is used to update the model weights and biases according to the gradient to reduce the prediction error. During the training process, the validation set is used regularly to evaluate the model performance until the model performance meets the requirements. During the model training process, the mean square error MSE algorithm is used to evaluate the performance of the deep foundation pit state prediction model. The specific algorithm is:
[0095]
[0096] Where n is the number of data points, Y pred,i is the temperature value predicted by the model, Y actual,i is the actual measured temperature value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.
[0097] S6: After training is completed, the weights and biases of the LSTM model are saved for subsequent model loading and prediction. The trained model is deployed for subsequent deep foundation pit state prediction, providing strong protection for deep foundation pit construction safety.
[0098] Before deploying the deep foundation pit state prediction model, a test data set that is completely different from the training set and validation set is constructed, and the deep foundation pit state prediction model is tested using the test data set. The final performance of the model is evaluated by calculating the prediction accuracy, recall rate, and F1 score. The specific method is as follows:
[0099] Load the trained deep foundation pit state prediction model and ensure that the model parameters are consistent with those during training. Input the test data set into the model, perform state prediction, and obtain the prediction results. Compare the prediction results with the actual state data in the test data set and calculate the prediction error.
[0100] To comprehensively evaluate the performance of the model, accuracy, recall, and F1 score are used as evaluation indicators. Accuracy measures the proportion of samples predicted correctly by the model to the total number of samples, reflecting the overall prediction accuracy of the model. Recall measures the proportion of positive samples predicted by the model to the actual number of positive samples, reflecting the model's ability to identify positive samples. The F1 score comprehensively considers accuracy and recall, calculates the harmonic mean of the two, and serves as a comprehensive evaluation indicator of model performance.
[0101] According to the model testing process, the prediction results of the test data set are obtained. According to the prediction results and the actual status data, the accuracy, recall rate and F1 score are calculated. The calculation results are analyzed to evaluate the prediction performance and generalization ability of the model. If the model performance meets the expected requirements, subsequent deployment is carried out; if it does not meet the requirements, the model is further optimized and adjusted.
[0102] The deep foundation pit state prediction model training method provided by the present invention can make full use of the time series processing capability of the LSTM model, explore the historical laws of deep foundation pit state changes, and realize accurate prediction of the future state of the deep foundation pit.
[0103] Embodiment 3:
[0104] The graphical interface of the user interaction component includes a real-time monitoring curve display area, a prediction curve display area, an alarm information display area and a user operation function area.
[0105] The real-time monitoring curve display area is responsible for displaying the monitoring data curves of various monitoring parameters of the deep foundation pit in real time. These data curves cover key indicators such as displacement, settlement, stress, groundwater level and soil pressure, and are transmitted by the real-time monitoring components. Through this area, users can clearly observe the changing trends of various monitoring parameters over time, so as to grasp the status of the deep foundation pit in real time.
[0106] The prediction curve display area is used to display the prediction curves of the state changes of each monitoring parameter of the deep foundation pit in the future. These prediction curves are output by the model calling component based on the prediction results of the LSTM model. Users can identify potential risks in advance by comparing the current state with the predicted state, providing strong support for taking preventive measures.
[0107] The alarm information display area is used to display the alarm information generated by the alarm judgment component in real time. When the monitoring parameters exceed the preset threshold, the alarm judgment component will immediately generate alarm information and display it in this area. The alarm information includes the monitoring parameters that exceed the preset threshold, the alarm level, and the recommended measures, providing users with timely and accurate alarm prompts.
[0108] The user operation function area provides users with a variety of interface operation functions to meet the different needs of users. Specifically including:
[0109] Zoom and drag the curve chart: Users can easily view detailed data and further analyze the status changes of deep foundation pits by zooming and dragging.
[0110] Select a specific time period for data analysis: Users can select a time period of interest and conduct in-depth analysis of the monitoring data within that time period to obtain more valuable information.
[0111] Set alarm thresholds: Users can flexibly set alarm thresholds according to actual needs to ensure the accuracy and sensitivity of the alarm system.
[0112] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0113] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A deep foundation pit automatic monitoring system, characterized by: It includes the perception end, data processing end, prediction end and monitoring end; The sensing end includes a variety of sensor components, including an inclinometer component, a settlement meter component, a pressure sensor component, a groundwater level meter component, and an earth pressure meter component; the sensing end is used to collect various parameters of the deep foundation pit and the surrounding environment in real time, including displacement, settlement, stress, groundwater level, and earth pressure, and transmit the collected data to the data processing end; The data processing end includes a data collection component and a data processing component; the data collection component is used to receive the original data transmitted by the sensing end, and the data processing component is used to perform data cleaning, denoising and solving on the received data; The prediction end includes a model building component, a feature extraction component, a model evaluation component and a model optimization component; the model building component is used to use machine learning technology to build a deep foundation pit state prediction model, which is based on historical monitoring data and deep foundation pit engineering data to predict the state changes of various monitoring parameters of the foundation pit in the future; the feature extraction component is used to select features that have an impact on the foundation pit state prediction from historical monitoring data and deep foundation pit engineering data as model input parameters; the model evaluation component is used to evaluate the performance of the trained deep foundation pit state prediction model; the model optimization component is used to regularly update and optimize the deep foundation pit state prediction model according to new monitoring data or model evaluation results to adapt to changes in the foundation pit construction process; The monitoring end includes a real-time monitoring component, a model calling component, an alarm judgment component and a user interaction component; the real-time monitoring component is used to draw the monitoring data transmitted by the data processing end into a real-time curve; the model calling component is used to call the deep foundation pit state prediction model of the prediction end, input the engineering data of the current deep foundation pit, and the model outputs the state changes of various monitoring parameters of the deep foundation pit in a period of time in the future. The model calling component further draws the results of the model output into a prediction curve; the alarm judgment component is used to compare the difference between the real-time curve and the prediction curve in real time, and combine the preset alarm threshold to determine whether the deep foundation pit is in a safe state. If the threshold is exceeded, an alarm message is generated; the user interaction component is used to provide a graphical interface to display the real-time monitoring curve output by the real-time monitoring component, the prediction curve output by the model calling component, and the alarm information generated by the alarm judgment component, and provide interface operation functions to the user.
2. The deep foundation pit automatic monitoring system according to claim 1 is characterized in that: The data cleaning of the data processing component includes processing missing values and outliers. For missing values, linear interpolation algorithm is used to fill them; for outliers, threshold judgment method is used to regard data points exceeding the preset threshold range as outliers and replace them with the mean of their neighboring points. The data processing component uses a sliding average filtering algorithm to perform data denoising, and the algorithm formula is: Among them, y[n] is the denoised data, x[n] is the original data, and N is the size of the sliding window.
3. The deep foundation pit automatic monitoring system according to claim 2 is characterized in that: The data processing component uses the least square method to perform data calculation to fit the relationship between the sensor data and the actual physical quantity. The algorithm formula is: θ=(X T X) -1 X T Y Among them, θ is the fitting parameter, X is the input data matrix, Y is the output data vector, X T is the transposed matrix of X, (X T X) -1 For X T The inverse matrix of X.
4. The deep foundation pit automatic monitoring system according to claim 3 is characterized in that: The feature extraction component evaluates the importance of features by calculating the correlation between each feature and the target variable; the linear correlation between each feature and the target variable is measured using the Pearson correlation coefficient, which is calculated as: Among them, r represents the correlation coefficient between the feature and the target variable, n represents the number of samples, x represents the feature value, and y represents the target variable value; According to the calculated correlation coefficient, the features whose absolute values are greater than the set threshold are selected as useful features to form a feature data set for the subsequent training of the deep foundation pit state prediction model.
5. The deep foundation pit automatic monitoring system according to claim 4 is characterized in that: The method of the model building component to train the deep foundation pit state prediction model is: S1: Collect historical data, which includes the status data of the deep foundation pit project at each monitoring node, including displacement, settlement, stress, groundwater level and soil pressure, and basic data corresponding to the deep foundation pit project, including the project's geographical location, geological conditions, design parameters and construction methods; S2: Standardize the cleaned feature data and filter out the features used for model training through the feature extraction component to form a training data set; S3: Divide the training data set into a training set and a validation set; S4: Select the long short-term memory network LSTM model to build a deep foundation pit state prediction model, and initialize the parameters of the model; the initialized parameters include: initializing the structural parameters of the LSTM model, including the feature dimension of the input layer, the number of hidden layers and the number of hidden units in each layer, and the number of nodes in the output layer, where the feature dimension of the input layer matches the number of extracted deep foundation pit state-related features, and the number of nodes in the output layer is equal to the predicted target number, that is, the state prediction value under all monitoring nodes; initialize the model weights and biases, and use small random numbers for initialization; set the activation function of the LSTM model, including setting the Sigmoid function for the gating mechanism, and setting the Tanh function for the calculation of candidate memory units and hidden states; configure the optimization algorithm, and set the learning rate parameters; S5. Model training and evaluation: Use the training set data to iteratively train the selected model. During the training process, the validation set is regularly used to evaluate the model performance until the model performance meets the requirements. The specific method is: Use the training set data to iteratively train the LSTM model. The training set data includes the actual status data of the deep foundation pit project at each monitoring node and the basic data of the deep foundation pit project. In each iteration, the following operations are performed: a) Send a batch of input data to the LSTM model, and update the memory unit and hidden state by calculating the values of the input gate, forget gate and output gate, and finally obtain the predicted output sequence, which is the state change curve of the deep foundation pit project at all monitoring nodes; b) Calculate the loss function value based on the predicted output and the true label; c) Calculate the gradient of the loss function with respect to the model parameters through the back propagation algorithm; d) Use the optimization algorithm to update the model weights and biases according to the gradient; S6: After training is completed, the weights and biases of the model are saved, and the model is deployed for subsequent deep foundation pit status prediction.
6. The deep foundation pit automatic monitoring system according to claim 5, characterized in that: During the model training process, the mean square error (MSE) algorithm is used to evaluate the performance of the deep foundation pit state prediction model. The specific algorithm is: Where n is the number of data points, Y pred,i is the temperature value predicted by the model, Y actual,i is the actual measured temperature value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.
7. The deep foundation pit automatic monitoring system according to claim 6 is characterized by: The graphical interface of the user interaction component includes a real-time monitoring curve display area, a prediction curve display area, an alarm information display area and a user operation function area; In the real-time monitoring curve display area, the monitoring data curves of various monitoring parameters of the deep foundation pit transmitted by the real-time monitoring component are displayed in real time, including displacement, settlement, stress, groundwater level and soil pressure; In the prediction curve display area, the prediction curve of the state change of each monitoring parameter of the deep foundation pit output by the model calling component in the future period is displayed, which is used for users to compare the current state with the predicted state and identify potential risks in advance; In the alarm information display area, the alarm information generated by the alarm judgment component is displayed in real time, including the monitoring parameters that exceed the preset threshold, the alarm level and the recommended measures; In the user operation function area, interface operation functions are provided, including zooming and dragging the curve chart to view detailed data, selecting a specific time period for data analysis, and setting alarm thresholds.
8. The deep foundation pit automatic monitoring system according to claim 7, characterized in that: Before deploying the deep foundation pit state prediction model, a test dataset that is completely different from the training set and validation set is constructed, and the deep foundation pit state prediction model is tested using the test dataset. The final performance of the model is evaluated by calculating the prediction accuracy, recall rate, and F1 score.
Citation Information
Cited By
Comprehensive experimental device for refrigeration and heat exchange of air conditioner
CN118913738A
Soil pressure analysis method and system for subway double-side extension foundation pit support structure
CN119358109A
Real-time stress analysis and early warning system for wharf breakwater
CN119416331A
Deep foundation pit monitoring parameter prediction method based on neural network
CN119441742A
Rail transit foundation pit support modeling method based on multi-source data fusion
CN120562009A
