High pier creeping formwork monitoring and early warning method and system based on machine learning

Abstract

The invention discloses a high-pier creeping formwork monitoring and early warning method and system based on machine learning, and the method comprises the steps: collecting data transmitted by a sensor disposed on a high-pier creeping formwork system, and carrying out the preprocessing of the collected sensor data; the method comprises the following steps: preprocessing sensor data, carrying out multi-source data fusion on the preprocessed sensor data, carrying out data multi-source fusion on the processed data, comprehensively considering the common influence of a plurality of types of sensors, training a fused value through a security level judgment model based on machine learning, outputting different alarm levels, and sending out alarm information. According to the invention, multi-dimensional and multi-angle safety monitoring of the creeping formwork system can be realized, the risk hidden danger of the creeping formwork system can be early warned in time, and the safety of high pier construction is effectively guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning and relates to a high pier climbing formwork monitoring and early warning method and system based on machine learning. Background Art

[0002] The high pier climbing formwork system is mainly used in the construction process of high bridge piers and other structures. By lifting the formwork and supporting structure of the climbing formwork system as a whole, continuous concrete pouring of high piers and other structures can be achieved. Due to the influence of the environment and construction loads, the stability of the high pier climbing formwork system is particularly important for the safety of high pier construction. At present, most people judge the stability of the climbing formwork system through manual experience. Although there is also a technology that uses sensor equipment to monitor the climbing formwork system, the deployed sensors are single and cannot monitor the safety status of the high pier climbing formwork system in real time and comprehensively, nor can they provide timely warnings for existing safety risks. Summary of the invention

[0003] The purpose of the present invention is to solve the problem that the sensor equipment in the prior art cannot monitor the safety status of the high pier climbing formwork system in real time and comprehensively, and there is no timely warning of the existing safety risks, and to provide a high pier climbing formwork monitoring and early warning method and system based on machine learning.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions: The high pier climbing formwork monitoring and early warning method based on machine learning includes: Collect data transmitted by sensors deployed on the high pier climbing formwork system, and pre-process the collected sensor data; Perform multi-source data fusion on the pre-processed sensor data to obtain the nonlinear relationship between environmental factors and creeping form deformation; Based on the nonlinear relationship between environmental factors and climbing formwork deformation and combined with Kalman filtering, the state of high pier climbing formwork is predicted; Based on the predicted state of the high pier climbing formwork, the fused data sequence is obtained; Divide the preprocessed sensor data and fusion data sequences into training sets, test sets, and validation sets; The sensor data and fusion data are grouped into several sample groups. The feature values ​​of each sample group in the training set and the validation set are obtained. The features corresponding to each sample group are grouped into a feature group. Each feature group corresponds to an alarm level. All feature groups constitute a complete feature set. and Téval; Build a sequential model, pass T_train and T_val to the sequential model for training, make predictions on the test set, and output an array representing the alarm level; The array based on the alarm level automatically triggers the alarm mechanism and generates alarm information of the corresponding level.

[0005] A further improvement of the present invention is: The preprocessing of the collected sensor data is specifically as follows: Determine the missing value position of the collected data, and use the known data points before and after the position as interpolation nodes; , the data collected by the corresponding sensor is , the time point of missing values Located in the interval Use formula (1) to construct the interval A cubic polynomial on :

[0006] in, are the coefficients of the polynomial, exist The function value at As an estimate of the missing values.

[0007] The preprocessing of the collected sensor data also includes: processing abnormal values ​​in various sensors based on moving smoothing replacement, specifically: Choose an odd window size , calculate the average value within the window; for each time point , determine a centralized window , the average value of all data points in the window is obtained by formula (2) , specifically:

[0008] Pre-set thresholds , when the difference between the original value and the average value in the window exceeds When the original value is considered to be an outlier, the average value is used Replace the outliers, and the time series data after preprocessing is recorded as .

[0009] The multi-source data fusion is performed on the pre-processed data to obtain the nonlinear relationship between the environmental factors and the creeping form deformation, specifically: Get the mean of preprocessed time series data and standard deviation ; Mean based on time series data and standard deviation , normalize each data point of the time series data; The normalized wind speed, temperature and humidity data are used as the input of the multi-layer perceptron MLP, and the complex nonlinear relationship between the environmental factors of wind speed, temperature and humidity data and the target variables is captured through multiple hidden layers and nonlinear activation functions of the multi-layer perceptron MLP.

[0010] The normalized wind speed, temperature and humidity data are used as the input of the multi-layer perceptron MLP, and the complex nonlinear relationship between the environmental factors of the wind speed, temperature and humidity data and the target variable is captured through multiple hidden layers and nonlinear activation functions of the multi-layer perceptron MLP, specifically: The number of nodes in the input layer of the multilayer perceptron is equal to the number of input features, namely the three variables of wind speed, temperature and humidity, so the input layer has 3 nodes; The multi-layer perceptron MLP comprises a plurality of hidden layers, each of which comprises a plurality of nodes; each node receives weighted inputs from all nodes in the previous layer, and applies an activation function to introduce nonlinearity; wherein, for the first nodes, whose weighted input Expressed as

[0011] in, is the output of the i-th node in the previous layer, It is from i The weight from the node to the jth node, is the bias term of the jth node, and n is the number of nodes in the previous layer; Through the ReLU activation function, the weighted input The output of the node is obtained by the activation function f ;

[0012] Among them, the activation function f is the ReLU activation function; The output layer of the multi-layer perceptron includes a node for outputting the nonlinear influence coefficient of environmental factors of wind speed, temperature and humidity data on the high pier climbing formwork; The multi-layer perceptron MLP is trained based on the feedback of actual monitoring data. During the training process, the weights and bias terms are adjusted through the back propagation algorithm to minimize the error between the predicted value and the actual value; its loss function is:

[0013] Where m is the number of samples, is the actual influence coefficient of the kth sample, It is the nonlinear influence coefficient of the kth sample after being processed by the multi-layer perceptron.

[0014] The actual state of the high pier climbing formwork is predicted and corrected based on the nonlinear relationship between environmental factors and the deformation of the climbing formwork and Kalman filtering, specifically: The nonlinear influence coefficient output by the multilayer perceptron is used as the prediction input, and the displacement and inclination data collected by the sensor are used as the observation input to initialize the state estimation and error covariance matrix of the Kalman filter; the state of the next time step is predicted through the Kalman filter; the state estimation and error covariance matrix are updated, and the updated state estimation is used as the error covariance matrix. As the output of Kalman filter; Perform Kalman filter state estimation prediction:

[0015] Among them, A is the state transfer matrix, B is the matrix that converts the multilayer perceptron output into the state space input, is the process noise; The error covariance matrix is ​​predicted as follows:

[0016] in, is the forecast error covariance, yes The posterior error covariance at time t, Q is the process noise covariance; The Kalman filter performs an update step, using the observed data to correct the predicted state; the Kalman gain is calculated using formula (8), Determines the weight between predictions and observations;

[0017] in, is the Kalman gain, H is the observation matrix, and R is the observation noise covariance, which includes the noise variance of displacement and inclination observations; The process of updating the state estimation is shown in formula (9):

[0018] in, is the posterior state estimate at time v, and the observation vector Contains both displacement and inclination data The process of updating the error covariance is shown in formula (10):

[0019] in, is the posterior error covariance at time v, and I is the identity matrix.

[0020] The sensor data and fusion data are grouped into several sample groups, the feature values ​​of each sample group in the training set and the validation set are obtained, the features corresponding to each sample group are grouped into a feature group, and all feature groups constitute a complete feature set. and T_val, specifically: The sensor data and fusion data at each moment form a sample group, which is recorded as ; Manually mark each sample group in the training set, set four alarm levels of "low", "medium", "high" and "safe", and each sample group corresponds to an alarm level; calculate the mean, variance, maximum value and minimum value of each sample group in the training set, and these four features corresponding to each sample group constitute a feature group, recorded as ,,Each feature group corresponds to a label, that is, the alarm level corresponding to the original sample group; all feature groups constitute a complete feature set and Téval.

[0021] The sequential model is constructed, T_train and T_val are passed to the sequential model for training, and the test set is predicted, and an array representing the alarm level is output, specifically: a sequential model is created, a first fully connected layer is added to the sequential model, including 64 neurons, and ReLU is used as an activation function; a second fully connected layer is added, including 32 neurons, and ReLU activation function is used again for activation, and finally an output layer is added, including 4 neurons, corresponding to the four levels predicted by the model; Configure the optimizer and loss function of the sequential model; select the Adam optimizer and mean square error MSE as the loss function; the MSE calculation formula is as follows (11):

[0022] in, Indicates The true value of the samples, represents the corresponding estimated value, represents the number of samples; Pass T_train and T_val to the sequential model, which will be trained based on these data, and continuously adjust the model parameters to minimize the loss function to obtain the most optimized sequential model. The test set is predicted through y_pred = model.predict(), where the model.predict function is used to predict the input data; the prediction of the test set outputs an array representing the alarm level.

[0023] The high pier climbing formwork monitoring and early warning system based on machine learning includes: A preprocessing module, wherein the preprocessing module collects data transmitted by sensors arranged on the high pier climbing formwork system and preprocesses the collected sensor data; A first acquisition module, which performs multi-source data fusion on the pre-processed sensor data to obtain a nonlinear relationship between environmental factors and creeping form deformation; A prediction module, which predicts the state of the high pier climbing formwork based on the nonlinear relationship between environmental factors and the deformation of the climbing formwork and combined with Kalman filtering; A second acquisition module, wherein the second acquisition module acquires a fused data sequence based on a predicted state of the high pier climbing formwork; A partitioning module, wherein the partitioning module partitions the preprocessed sensor data and fusion data sequence to obtain a training set, a test set, and a validation set; The third acquisition module groups the sensor data and the fusion data into several sample groups, obtains the feature values ​​of each sample group in the training set and the validation set, and forms a feature group with the features corresponding to each sample group. Each feature group corresponds to an alarm level, and all feature groups form a complete feature set. and Téval; A training module, wherein the training module constructs a sequential model, passes T_train and T_val to the sequential model for training, and predicts the test set to output an array representing the alarm level; A trigger module automatically triggers an alarm mechanism based on an array of alarm levels and generates alarm information of corresponding levels.

[0024] Compared with the prior art, the present invention has the following beneficial effects: The present invention collects data transmitted by sensors and pre-processes abnormal values ​​and missing values ​​in the data; performs multi-source data fusion on the processed data, comprehensively considers the common influence of multiple categories of sensors, trains the fused values ​​through a safety level judgment model based on machine learning, outputs different alarm levels, and issues alarm information, so that managers can take corresponding safety measures in advance. The present invention can realize multi-dimensional and multi-angle safety monitoring of the climbing formwork system, timely warn of the risks and hidden dangers of the climbing formwork system, and effectively ensure the safety of high pier construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 A schematic diagram of a process of a high pier climbing formwork monitoring and early warning method based on machine learning of the present invention; Figure 2 It is a structural schematic diagram of the high pier climbing formwork monitoring and early warning system based on machine learning of the present invention; Figure 3 Another flow chart of the high pier climbing formwork monitoring and early warning method based on machine learning of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0028] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. 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.

[0029] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0030] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0031] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0032] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0033] The present invention is further described in detail below in conjunction with the accompanying drawings: See also Figures 1 to 3 The present invention discloses a high pier climbing formwork monitoring and early warning method based on machine learning, comprising: S101, collecting data transmitted by sensors arranged on the high pier climbing formwork system, and preprocessing the collected sensor data; Determine the missing value position of the collected data, and use the known data points before and after the position as interpolation nodes; , the data collected by the corresponding sensor is , the time point of missing values Located in the interval Use formula (1) to construct the interval A cubic polynomial on :

[0034] in, are the coefficients of the polynomial, exist The function value at As an estimate of the missing values.

[0035] The collected sensor data is preprocessed, and the process also includes: processing abnormal values ​​in various sensors based on moving smoothing replacement, specifically: Choose an odd window size , calculate the average value within the window; for each time point , determine a centralized window , the average value of all data points in the window is obtained by formula (2) , specifically:

[0036] Pre-set thresholds , when the difference between the original value and the average value in the window exceeds When the original value is considered to be an outlier, the average value is used Replace the outliers, and the time series data after preprocessing is recorded as .

[0037] S102, performing multi-source data fusion on the pre-processed sensor data to obtain the nonlinear relationship between the environmental factors and the creeping form deformation; Get the mean of preprocessed time series data and standard deviation ; Mean based on time series data and standard deviation , normalize each data point of the time series data; The normalized wind speed, temperature and humidity data are used as the input of the multi-layer perceptron MLP, and the complex nonlinear relationship between the environmental factors of wind speed, temperature and humidity data and the target variables is captured through multiple hidden layers and nonlinear activation functions of the multi-layer perceptron MLP.

[0038] The number of nodes in the input layer of the multilayer perceptron is equal to the number of input features, namely the three variables of wind speed, temperature and humidity, so the input layer has 3 nodes; The multi-layer perceptron MLP comprises a plurality of hidden layers, each of which comprises a plurality of nodes; each node receives weighted inputs from all nodes in the previous layer, and applies an activation function to introduce nonlinearity; wherein, for the first nodes, whose weighted input Expressed as

[0039] in, is the output of the i-th node in the previous layer, is the weight from the i-th node to the j-th node, is the bias term of the jth node, and n is the number of nodes in the previous layer; Through the ReLU activation function, the weighted input The output of the node is obtained by the activation function f ;

[0040] Among them, the activation function f is the ReLU activation function; The output layer of the multi-layer perceptron includes a node for outputting the nonlinear influence coefficient of environmental factors of wind speed, temperature and humidity data on the high pier climbing formwork; The multi-layer perceptron MLP is trained based on the feedback of actual monitoring data. During the training process, the weights and bias terms are adjusted through the back propagation algorithm to minimize the error between the predicted value and the actual value; its loss function is:

[0041] Where m is the number of samples, is the actual influence coefficient of the kth sample, It is the nonlinear influence coefficient of the kth sample after being processed by the multi-layer perceptron.

[0042] S103, predicting the state of the high pier climbing formwork based on the nonlinear relationship between environmental factors and climbing formwork deformation and Kalman filtering; The nonlinear influence coefficient output by the multilayer perceptron is used as the prediction input, and the displacement and inclination data collected by the sensor are used as the observation input to initialize the state estimation and error covariance matrix of the Kalman filter; the state of the next time step is predicted through the Kalman filter; the state estimation and error covariance matrix are updated, and the updated state estimation is used as the error covariance matrix. As the output of Kalman filter; Perform Kalman filter state estimation prediction:

[0043] Among them, A is the state transfer matrix, B is the matrix that converts the multilayer perceptron output into the state space input, is the process noise; The error covariance matrix is ​​predicted as follows:

[0044] in, is the forecast error covariance, yes is the a posteriori error covariance at time t, Q is the process noise covariance; The Kalman filter performs an update step, using the observed data to correct the predicted state; the Kalman gain is calculated using formula (8), Determines the weight between predictions and observations;

[0045] in, is the Kalman gain, H is the observation matrix, and R is the observation noise covariance, which includes the noise variance of displacement and inclination observations; The process of updating the state estimation is shown in formula (9):

[0046] in, is the posterior state estimate at time v, and the observation vector Contains both displacement and inclination data The process of updating the error covariance is shown in formula (10):

[0047] in, is the posterior error covariance at time v, and I is the identity matrix.

[0048] S104, acquiring a fused data sequence based on the predicted state of the high pier climbing formwork; S105, dividing the preprocessed sensor data and fusion data sequence into a training set, a test set, and a validation set; S106, the sensor data and the fusion data are grouped into several sample groups, the feature values ​​of each sample group in the training set and the validation set are obtained, the features corresponding to each sample group are grouped into a feature group, each feature group corresponds to an alarm level, and all feature groups constitute a complete feature set and Téval; The first The sensor data and fusion data at each moment form a sample group, which is recorded as ; Manually mark each sample group in the training set, set four alarm levels of "low", "medium", "high" and "safe", and each sample group corresponds to an alarm level; calculate the mean, variance, maximum value and minimum value of each sample group in the training set, and these four features corresponding to each sample group constitute a feature group, recorded as ,,Each feature group corresponds to a label, that is, the alarm level corresponding to the original sample group; all feature groups constitute a complete feature set and Téval.

[0049] S107, construct a sequential model, pass T_train and T_val to the sequential model for training, predict the test set, and output an array representing the alarm level; Create a sequential model and add the first fully connected layer to the sequential model, including 64 neurons, using ReLU as the activation function; add a second fully connected layer with 32 neurons, again activated with the ReLU activation function, and finally add an output layer with 4 neurons, corresponding to the four levels predicted by the model; Configure the optimizer and loss function of the sequential model; select the Adam optimizer and mean square error MSE as the loss function; the MSE calculation formula is as follows (11):

[0050] in, Indicates The true value of the samples, represents the corresponding estimated value, represents the number of samples; Pass T_train and T_val to the sequential model, which will be trained based on these data, and continuously adjust the model parameters to minimize the loss function to obtain the most optimized sequential model. The test set is predicted through y_pred = model.predict(), where the model.predict function is used to predict the input data; the prediction of the test set outputs an array representing the alarm level.

[0051] S108, automatically triggering an alarm mechanism based on the array of alarm levels to generate alarm information of corresponding levels.

[0052] See also Figure 2 The present invention discloses a high pier climbing formwork monitoring and early warning system based on machine learning, comprising: A preprocessing module, wherein the preprocessing module collects data transmitted by sensors arranged on the high pier climbing formwork system and preprocesses the collected sensor data; A first acquisition module, which performs multi-source data fusion on the pre-processed sensor data to obtain a nonlinear relationship between environmental factors and creeping form deformation; A prediction module, which predicts the state of the high pier climbing formwork based on the nonlinear relationship between environmental factors and the deformation of the climbing formwork and Kalman filtering; A second acquisition module, wherein the second acquisition module acquires a fused data sequence based on a predicted state of the high pier climbing formwork; A partitioning module, wherein the partitioning module partitions the preprocessed sensor data and fusion data sequence to obtain a training set, a test set, and a validation set; The third acquisition module groups the sensor data and the fusion data into several sample groups, obtains the feature values ​​of each sample group in the training set and the validation set, and forms a feature group with the features corresponding to each sample group. Each feature group corresponds to an alarm level, and all feature groups form a complete feature set. and Téval; A training module, wherein the training module constructs a sequential model, passes T_train and T_val to the sequential model for training, and predicts the test set to output an array representing the alarm level; A trigger module automatically triggers an alarm mechanism based on an array of alarm levels and generates alarm information of corresponding levels.

[0053] Example: See also Figure 3 The present invention discloses a high pier climbing formwork monitoring and early warning method based on machine learning, comprising: Step 1: Sensor data collection and preprocessing: In order to monitor the status of the high pier climbing formwork system, laser rangefinders, inclination sensors, wind speed sensors, and temperature and humidity sensors are deployed on the high pier climbing formwork. Due to the influence of objective factors such as the environment and weather, the data collected by the sensor equipment may contain abnormalities and missing data. Therefore, it is necessary to first eliminate abnormal data and fill in missing data. This preprocessing process is designed to ensure data quality, reduce errors and uncertainties in subsequent analysis, thereby improving the reliability and effectiveness of the data, and laying a solid foundation for subsequent data analysis and modeling of the system.

[0054] Step 2: Multi-source data fusion: For the data collected by sensor equipment, it is necessary to use multi-source data fusion technology to achieve all-round monitoring of the high pier climbing formwork system. In this processing process, the multi-layer perceptron technology is mainly used. The multi-layer perceptron can capture the complex nonlinear relationship between environmental factors and the deformation of the climbing formwork, quantify these influencing factors and use them as the disturbance value input of the Kalman filter, so as to more accurately predict and correct the actual state of the high pier climbing formwork. The prediction model analyzes the environmental data through the multi-layer perceptron to provide a preliminary estimate of the state of the high pier climbing formwork system, while the observation data is the current state of the high pier climbing formwork system obtained in real time from the displacement and inclination sensors. These data are processed by the Kalman filter to dynamically update the state of the high pier climbing formwork system, thereby improving the accuracy and reliability of state estimation.

[0055] Step 3: Alarm level determination based on machine learning: In the high pier climbing formwork safety monitoring system, the alarm level judgment based on machine learning is set to improve monitoring accuracy, enhance safety response speed and optimize management efficiency. The core function of this part is to intelligently analyze the monitoring data through machine learning algorithms, automatically identify abnormal conditions and accurately judge their danger levels, thereby triggering the corresponding level of alarm mechanism. First of all, this process improves the intelligence level of safety monitoring. The alarm system based on machine learning can also achieve all-weather, uninterrupted monitoring, effectively overcome the limitations and shortcomings of manual monitoring, and reduce the errors and subjectivity of manual judgment; secondly, fast and accurate alarm level judgment can help to take necessary measures in time, so as to carry out on-site emergency disposal and prevent the occurrence of safety accidents.

[0056] Step 4: Smart Security Alarm: After completing the alarm level judgment based on machine learning, the system will automatically trigger the corresponding alarm mechanism according to different alarm levels, generate alarm information of the corresponding level, and promptly release it to on-site staff or managers. The machine learning algorithm can quickly and accurately identify anomalies and judge their danger level, and the intelligent safety alarm will promptly issue an alarm based on this judgment, ensuring that managers can take appropriate emergency measures according to the specific alarm level. This linkage mechanism greatly improves the response speed and accuracy of the system, allowing managers to effectively intervene before potential risks deteriorate, avoid accidents, and ultimately improve the efficiency and response capabilities of overall safety management.

[0057] Furthermore, step 1 specifically includes the following steps: 1.1 Cubic spline interpolation is used to handle missing values ​​received by various sensors.

[0058] Furthermore, the time series data collected by various sensors are obtained. The time series data collected by the laser rangefinder is expressed as , the time series data collected by the tilt sensor is expressed as , the time series data collected by the wind speed sensor is expressed as , the time series data collected by the temperature sensor is represented as , the time series data collected by the humidity sensor is represented as The steps and operations for data preprocessing and normalization of time series data collected by various sensors are the same. The following takes the time series data collected by the laser rangefinder as an example, which is recorded as ,in, is the number of data points.

[0059] Furthermore, the missing value position that needs to be interpolated is determined, and the known data points before and after the position are used as interpolation nodes. , the corresponding data collected by the laser rangefinder is , the time point of missing values Located in the interval In the interval, use formula (1) to construct A cubic polynomial on (1) in, are the coefficients of the polynomial, which need to be determined based on the following four conditions (2) (3) (4) (5) Calculate exist The function value at As an estimate of the missing values.

[0060] 1.2 Use moving smoothing replacement to handle outliers in various sensor types.

[0061] Furthermore, choose an odd window size , calculate the average value within this window. For each time point , determine a centralized window , use formula (6) to calculate the average value of all data points in the window .

[0062] (6) Furthermore, we set a threshold , when the difference between the original value and the mean value exceeds When the original value is considered to be an outlier, the average value is used Replace outliers.

[0063] The time series data after preprocessing is recorded as .

[0064] Furthermore, step 2 specifically includes the following steps: 2.1 Normalize various sensor data.

[0065] Furthermore, equations (7) and (8) are used to calculate the mean of the time series data after preprocessing: and standard deviation .

[0066] (7) in, is the value of each data point after preprocessing.

[0067] (8) The standard deviation measures how far the data points deviate from the mean. Here we use the sample standard deviation, which is divided by , and obtain an unbiased estimate.

[0068] Furthermore, the Z score of each data point is calculated using formula (9).

[0069] (9) The Z-score represents the deviation of a data point from the mean, divided by the standard deviation, to produce a unitless measure. Replace each data point in the original data set with the calculated Z-score.

[0070] The time series data after normalization is recorded as .

[0071] 2.2 Use the multi-layer perceptron method to perform nonlinear processing on wind speed, temperature and humidity.

[0072] Furthermore, the normalized wind speed, temperature and humidity data are used as the input of the multi-layer perceptron. The multi-layer perceptron MLP is a feed-forward artificial neural network model that can capture the complex nonlinear relationship between environmental factors such as wind speed, temperature and humidity and the target variables through its multiple hidden layers and nonlinear activation functions. The number of nodes in the input layer of the multi-layer perceptron is equal to the number of input features, namely the three variables of wind speed, temperature and humidity, so the input layer has 3 nodes.

[0073] Furthermore, the multilayer perceptron contains multiple hidden layers, each of which contains a certain number of nodes. Each node receives weighted inputs from all nodes in the previous layer and applies an activation function to introduce nonlinearity. nodes, whose weighted input It can be expressed as (10) in, is the output of the ith node in the previous layer (the input layer or another hidden layer), is the weight from the i-th node to the j-th node, is the bias term of the jth node, and n is the number of nodes in the previous layer.

[0074] The activation function selects the ReLU function and the weighted input The output of the node is obtained by the activation function f .

[0075] (11) Furthermore, the output layer of the multilayer perceptron includes a node for outputting the nonlinear influence coefficient on the high pier climbing formwork.

[0076] Furthermore, the multilayer perceptron is trained based on the feedback of actual monitoring data. During the training process, the weights and bias terms are adjusted through the back propagation algorithm to minimize the error between the predicted value and the actual value. Its loss function is expressed as formula (12): (12) Where m is the number of samples, is the actual influence coefficient of the kth sample, It is the nonlinear influence coefficient of the kth sample after being processed by the multi-layer perceptron.

[0077] Furthermore, the reliability of the model is evaluated by using data that it has never seen as input to ensure that the model can also show good generalization ability when facing different data sets. The trained multi-layer perceptron model is integrated into the high pier climbing formwork safety monitoring system, which receives environmental data such as wind speed, temperature and humidity in real time, and outputs the nonlinear influence coefficient on the high pier climbing formwork for subsequent Kalman filtering.

[0078] 2.3 Apply Kalman filtering to realize the fusion estimation of the state of high pier climbing formwork.

[0079] Furthermore, the nonlinear influence coefficient output by the multilayer perceptron is used as the prediction input, and the displacement and inclination data collected by the displacement sensor and the inclination sensor are used as the observation input and passed into the Kalman filter model.

[0080] Furthermore, the state estimation and error covariance matrix of the Kalman filter are initialized. These initial values ​​are determined based on prior knowledge of the system or experimental data.

[0081] Furthermore, the Kalman filter performs a prediction step and uses the system's dynamic model to predict the state of the next time step. First, the state is predicted according to equation (13): (13) Among them, A is the state transfer matrix, B is the matrix that converts the multilayer perceptron output into the state space input, is the process noise.

[0082] Then, the error covariance is predicted according to formula (14): (14) in, is the forecast error covariance, yes is the a posteriori error covariance at time t, and Q is the process noise covariance.

[0083] Furthermore, the Kalman filter performs an update step and uses the observed data to correct the predicted state. The Kalman gain is calculated using equation (15): Determines the weight between predictions and observations.

[0084] (15) in, is the Kalman gain, H is the observation matrix, and R is the observation noise covariance, which includes the noise variance of displacement and inclination observations.

[0085] The process of updating the state estimation is as follows: (16) in, is the posterior state estimate at time v, and the observation vector Contains both displacement and inclination data The process of updating the error covariance is as follows: (17) in, is the posterior error covariance at time v, and I is the identity matrix.

[0086] Furthermore, the updated state estimate The output of Kalman filtering includes the displacement, inclination and other status information of the pier.

[0087] Furthermore, step 3 specifically includes the following steps: 3.1 Data preparation

[0088] According to step 1 and step 2, a set of fused data sequences is obtained .

[0089] 3.2 Divide the dataset.

[0090] Furthermore, the preprocessed sensor data are divided into a 70% training set, a 20% validation set and a 10% test set; wherein the training set and the validation set include the displacement data of the high pier climbing formwork system collected from the laser rangefinder, the inclination data of the high pier climbing formwork system collected from the inclination sensor, the wind speed data of the construction site collected from the wind speed sensor, the temperature and humidity data of the construction site collected from the temperature and humidity sensor, and the fusion data obtained from step 1 and step 2; the test set consists of the displacement data of the high pier climbing formwork system collected from the laser rangefinder, the inclination data of the high pier climbing formwork system collected from the inclination sensor, the wind speed data of the construction site collected from the wind speed sensor, and the temperature and humidity data of the construction site collected from the temperature and humidity sensor.

[0091] Furthermore, i The displacement data, inclination data, wind speed data, temperature and humidity data and fusion data at each moment constitute a sample group, which is recorded as . Based on expert experience and industry regulations, each sample group in the training set and validation set is manually labeled, and four alarm levels of "low", "medium", "high" and "safe" are set, with each sample group corresponding to one alarm level.

[0092] 3.3 Feature selection.

[0093] Furthermore, the four features of each sample group in the training set and the validation set are calculated, namely, the mean, variance, maximum value, and minimum value. The four features corresponding to each sample group constitute a feature group, which is recorded as , each feature group corresponds to a label, that is, the alarm level corresponding to the original sample group. All feature groups constitute a complete feature set and T_val; Specifically, the four features of mean, variance, maximum value, and minimum value are merged into a data table, recorded as DataFrame, and a row of data table containing these features is created. The features are added to the original DataFrame, and SelectKBest is used for feature selection. SelectKBest is a feature selection method that is mainly used to select the features with the most predictive ability. The first k best performing features are selected from the four features. Instantiate SelectKBest as selector= SelectKBest(f_classif, k=4), and use f_classif as the scoring function. f_classif is also called F test classification. It measures the ability to distinguish categories by calculating the F value of each feature. The higher the F value, the greater the contribution of the feature to distinguishing different categories. k=4 means selecting the first four best features. In order to fit and transform the selector, the selector.fit_transform method is used to select the features of mean, variance, maximum value, and minimum value, and return new data sets T_train and T_val containing only these features. The selector.fit_transform method consists of two steps: (1) fitting: calculating the importance or score of the features; (2) transforming: selecting the most important features based on the calculation results and transforming the dataset into a form containing only these features.

[0094] 3.4 Model construction and training.

[0095] Furthermore, a sequential model is created, adding the first fully connected layer with 64 neurons and using ReLU as the activation function to learn complex nonlinear relationships. Next, a second fully connected layer with 32 neurons is added, also using the ReLU activation function. Finally, an output layer is added with 4 neurons, corresponding to the four alarm levels predicted by the model.

[0096] Furthermore, configure the optimizer and loss function of the sequential model; select the Adam optimizer and mean square error MSE as the loss function. The MSE calculation formula is as follows (18):

[0097] in, Indicates The true value of the samples, represents the corresponding estimated value, Indicates the sample size.

[0098] Furthermore, T_train and T_val are passed to the sequential model to evaluate the performance changes of the model during the training process. The sequential model will be trained based on these data, and the model parameters will be continuously adjusted to minimize the loss function to obtain the optimized sequential model.

[0099] 3.5 Model prediction

[0100] Furthermore, the test set is predicted using y_pred = model.predict(), where the model.predict function is used to predict the input data. The test set is used as the input of the model.predict function to evaluate the generalization ability of the model, that is, the performance of the model on new data. Finally, the test set is predicted to output an array representing the alarm level.

[0101] Furthermore, step 4 specifically includes the following steps: 4.1 Monitoring alarm For each sample group, according to its safety risk status, there is an array representing the alarm level. For example, if the alarm level corresponding to the sample group is "medium", then in the array, the first dimension is 0, the second dimension is 1, the third dimension is 0, and the fourth dimension is 0. The array obtained by the above steps, where "high" represents the first-level warning: emergency level, indicates that the high pier climbing formwork system determines that there is a serious safety risk and needs to take emergency measures immediately. "Medium" represents the second-level warning: serious warning level, indicates that the high pier climbing formwork system determines that there is a large safety risk and needs to pay attention and take measures immediately, such as stopping work, overhauling, adjusting load, etc., "low" represents the third-level warning: warning level, indicates that the high pier climbing formwork system determines that there is a slight safety risk. "Safe" means that the high pier climbing formwork system is running normally and there is no risk.

[0102] 4.2 Alarm log records Alarm log recording is essential for the performance analysis and improvement of the high pier climbing formwork intelligent safety monitoring system, including: (1) Detailed description of the event: Record the detailed information of each alarm event on the safety status of the high pier climbing formwork system, including the duration and solution. The duration should describe the entire time process from the alarm signal being triggered to the resolution. The solution should clearly state what measures were taken to deal with the alarm event.

[0103] (2) Multi-level classification: The safety status alarm events of the high pier climbing formwork system can be divided into different levels according to the degree of urgency, which helps to better analyze and manage the safety risks of the high pier climbing formwork system. The levels can be divided into four categories: the first level warning represents the emergency level, the second level warning represents the serious warning level, the third level warning represents the warning level, and the fourth level represents safety. According to these levels, construction personnel can respond quickly and take corresponding measures to reduce or eliminate potential safety risks.

[0104] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A high pier climbing formwork monitoring and early warning method based on machine learning, characterized in that: include: Collect data transmitted by sensors deployed on the high pier climbing formwork system, and pre-process the collected sensor data; Perform multi-source data fusion on the pre-processed sensor data to obtain the nonlinear relationship between environmental factors and creeping form deformation; Based on the nonlinear relationship between environmental factors and climbing formwork deformation and combined with Kalman filtering, the state of high pier climbing formwork is predicted; Based on the predicted state of the high pier climbing formwork, the fused data sequence is obtained; Divide the preprocessed sensor data and fusion data sequences into training sets, test sets, and validation sets; The sensor data and fusion data are grouped into several sample groups. The feature values ​​of each sample group in the training set and the validation set are obtained. The features corresponding to each sample group are grouped into a feature group. Each feature group corresponds to an alarm level. All feature groups constitute a complete feature set. and Téval; Build a sequential model, pass T_train and T_val to the sequential model for training, make predictions on the test set, and output an array representing the alarm level; The array based on the alarm level automatically triggers the alarm mechanism and generates alarm information of the corresponding level.

2. The high pier climbing formwork monitoring and early warning method based on machine learning according to claim 1 is characterized in that: The preprocessing of the collected sensor data is specifically as follows: Determine the missing value position of the collected data, and use the known data points before and after the position as interpolation nodes; , the data collected by the corresponding sensor is , the time point of the missing value Located in the interval Use formula (1) to construct the interval A cubic polynomial on : in, are the coefficients of the polynomial, exist The function value at As an estimate of the missing values.

3. The high pier climbing formwork monitoring and early warning method based on machine learning according to claim 2 is characterized in that: The preprocessing of the collected sensor data also includes: processing abnormal values ​​in various sensors based on moving smoothing replacement, specifically: Choose an odd window size , calculate the average value within the window; for each time point , determine a centralized window , the average value of all data points in the window is obtained by formula (2) , specifically: Pre-set thresholds , when the difference between the original value and the average value in the window exceeds When the original value is considered to be an outlier, the average value is used Replace the outliers, and the time series data after preprocessing is recorded as .

4. The high pier climbing formwork monitoring and early warning method based on machine learning according to claim 3 is characterized in that: The multi-source data fusion is performed on the pre-processed data to obtain the nonlinear relationship between the environmental factors and the creeping form deformation, specifically: Get the mean of preprocessed time series data and standard deviation ; Mean based on time series data and standard deviation , normalize each data point of the time series data; The normalized wind speed, temperature and humidity data are used as the input of the multi-layer perceptron MLP, and the complex nonlinear relationship between the environmental factors of wind speed, temperature and humidity data and the target variables is captured through multiple hidden layers and nonlinear activation functions of the multi-layer perceptron MLP.

5. The high pier climbing formwork monitoring and early warning method based on machine learning according to claim 4 is characterized in that: The normalized wind speed, temperature and humidity data are used as the input of the multi-layer perceptron MLP, and the complex nonlinear relationship between the environmental factors of the wind speed, temperature and humidity data and the target variable is captured through multiple hidden layers and nonlinear activation functions of the multi-layer perceptron MLP, specifically: The number of nodes in the input layer of the multilayer perceptron is equal to the number of input features, namely the three variables of wind speed, temperature and humidity, so the input layer has 3 nodes; The multi-layer perceptron MLP comprises a plurality of hidden layers, each of which comprises a plurality of nodes; each node receives weighted inputs from all nodes in the previous layer, and applies an activation function to introduce nonlinearity; wherein, for the first nodes, whose weighted input Expressed as in, is the output of the i-th node in the previous layer, It is from i The weight from the node to the jth node, is the bias term of the jth node, and n is the number of nodes in the previous layer; Through the ReLU activation function, the weighted input The output of the node is obtained by the activation function f ; Among them, the activation function f is the ReLU activation function; The output layer of the multi-layer perceptron includes a node for outputting the nonlinear influence coefficient of environmental factors of wind speed, temperature and humidity data on the high pier climbing formwork; The multi-layer perceptron MLP is trained based on the feedback of actual monitoring data. During the training process, the weights and bias terms are adjusted through the back propagation algorithm to minimize the error between the predicted value and the actual value; its loss function is: Where m is the number of samples, is the actual influence coefficient of the kth sample, It is the nonlinear influence coefficient of the kth sample after being processed by the multi-layer perceptron.

6. The high pier climbing formwork monitoring and early warning method based on machine learning according to claim 5 is characterized in that: The actual state of the high pier climbing formwork is predicted and corrected based on the nonlinear relationship between environmental factors and the deformation of the climbing formwork and Kalman filtering, specifically: The nonlinear influence coefficient output by the multilayer perceptron is used as the prediction input, and the displacement and inclination data collected by the sensor are used as the observation input to initialize the state estimation and error covariance matrix of the Kalman filter; Use Kalman filtering to predict the state of the next time step; Update the state estimate and error covariance matrix, and update the state estimate As the output of Kalman filter; Perform Kalman filter state estimation prediction: Among them, A is the state transfer matrix, B is the matrix that converts the multilayer perceptron output into the state space input, is the process noise; The error covariance matrix is ​​predicted as follows: in, is the forecast error covariance, yes The posterior error covariance at time t, Q is the process noise covariance; The Kalman filter performs an update step, using the observed data to correct the predicted state; the Kalman gain is calculated using formula (8), Determines the weight between predictions and observations; in, is the Kalman gain, H is the observation matrix, and R is the observation noise covariance, which includes the noise variance of displacement and inclination observations; The process of updating the state estimation is shown in formula (9): in, is the posterior state estimate at time v, and the observation vector Contains both displacement and inclination data The process of updating the error covariance is shown in formula (10): in, is the posterior error covariance at time v, and I is the identity matrix.

7. The high pier climbing formwork monitoring and early warning method based on machine learning according to claim 6 is characterized in that: The sensor data and fusion data are grouped into several sample groups, the feature values ​​of each sample group in the training set and the validation set are obtained, the features corresponding to each sample group are grouped into a feature group, and all feature groups constitute a complete feature set. and T_val, specifically: The sensor data and fusion data at each moment form a sample group, which is recorded as ; Manually mark each sample group in the training set, set four alarm levels of "low", "medium", "high", and "safe", and each sample group corresponds to an alarm level; calculate the mean, variance, maximum value, and minimum value of each sample group in the training set. The four features corresponding to each sample group constitute a feature group, recorded as ,,Each feature group corresponds to a label, that is, the alarm level corresponding to the original sample group; all feature groups constitute a complete feature set and Téval.

8. The high pier climbing formwork monitoring and early warning method based on machine learning according to claim 7 is characterized in that: The sequential model is constructed, T_train and T_val are passed to the sequential model for training, and the test set is predicted, and an array representing the alarm level is output, specifically: a sequential model is created, a first fully connected layer is added to the sequential model, including 64 neurons, and ReLU is used as an activation function; a second fully connected layer is added, including 32 neurons, and ReLU activation function is used again for activation, and finally an output layer is added, including 4 neurons, corresponding to the four levels predicted by the model; Configure the optimizer and loss function of the sequential model; select the Adam optimizer and mean square error MSE as the loss function; the MSE calculation formula is as follows (11): in, Indicates The true value of the samples, represents the corresponding estimated value, represents the number of samples; Pass T_train and T_val to the sequential model, which will be trained based on these data, and continuously adjust the model parameters to minimize the loss function to obtain the most optimized sequential model. The test set is predicted through y_pred = model.predict(), where the model.predict function is used to predict the input data; the prediction of the test set outputs an array representing the alarm level.

9. The high pier climbing formwork monitoring and early warning system based on machine learning is characterized by: include: A preprocessing module, wherein the preprocessing module collects data transmitted by sensors arranged on the high pier climbing formwork system and preprocesses the collected sensor data; A first acquisition module, which performs multi-source data fusion on the pre-processed sensor data to obtain a nonlinear relationship between environmental factors and creeping form deformation; A prediction module, which predicts the state of the high pier climbing formwork based on the nonlinear relationship between environmental factors and the deformation of the climbing formwork and combined with Kalman filtering; A second acquisition module, wherein the second acquisition module acquires a fused data sequence based on a predicted state of the high pier climbing formwork; A partitioning module, wherein the partitioning module partitions the preprocessed sensor data and fusion data sequence to obtain a training set, a test set, and a validation set; The third acquisition module groups the sensor data and the fusion data into several sample groups, obtains the feature values ​​of each sample group in the training set and the validation set, and forms a feature group with the features corresponding to each sample group. Each feature group corresponds to an alarm level, and all feature groups form a complete feature set. and Téval; A training module, wherein the training module constructs a sequential model, passes T_train and T_val to the sequential model for training, and predicts the test set to output an array representing the alarm level; A trigger module automatically triggers an alarm mechanism based on an array of alarm levels and generates alarm information of corresponding levels.

Images