A method for online monitoring of slump of cast agitated concrete

Abstract

This application provides a method for online monitoring of the slump of poured concrete, comprising: deploying several concrete surface moisture sensors and wind speed sensors at the concrete pouring site, presetting the sensor sampling frequency, measuring the concrete surface and environment within a preset range, and obtaining concrete surface moisture evaporation rate data and wind speed data; constructing prediction models for different wind speed modes based on statistical characteristics of different wind speed modes, using the statistical characteristics of wind speed modes as input and the statistical characteristics of evaporation rate under the corresponding wind speed modes as output; after newly collected wind speed data, determining the wind speed mode corresponding to the newly collected wind speed data, and obtaining the predicted value of concrete surface moisture evaporation rate according to the prediction model under the corresponding wind speed mode.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for online monitoring of the slump of poured and mixed concrete. Background Technology

[0002] Concrete slump is a crucial indicator of concrete workability, reflecting its plasticizing and pumpability properties, and is essential for ensuring smooth construction. Evaporation of moisture from the concrete surface directly reduces the free water content in the concrete clinker, leading to slump loss. This is especially true in the initial stages of pouring; excessively rapid evaporation can cause surface dehydration, resulting in plastic shrinkage cracks, further reducing slump and affecting workability. Conversely, proper moisture retention helps maintain slump, ensuring good workability and final strength. Environmental factors such as humidity, wind speed, and temperature all influence the rate of moisture evaporation from concrete. Therefore, real-time monitoring of surface moisture evaporation rate and wind speed fluctuations is critical at concrete pouring sites. However, wind speeds in the natural environment exhibit complex and unstable patterns, such as gusts and turbulence, and the relationship between these patterns and the rate of moisture evaporation remains unclear. Accurately identifying these unstable wind speed patterns and correlating them with evaporation rate data presents a pressing technical challenge. Specifically, an algorithm needs to be developed to extract high-frequency and low-frequency components from continuous natural wind speed time series and determine the occurrence time and duration of gusts and turbulence based on preset thresholds. Simultaneously, a model needs to be established to link the statistical characteristics of different natural wind speed patterns with the evaporation rate of the corresponding time periods, enabling evaporation rate prediction. To effectively address this problem, research not only needs to delve into signal processing and pattern recognition but also overcome technical obstacles in time synchronization and capturing short-term wind speed changes. Furthermore, ensuring the accuracy and real-time nature of sensor-collected data in complex construction site environments is also a crucial factor to consider. The ultimate goal is to dynamically adjust construction strategies through an optimized monitoring system to maintain the ideal slump of concrete, ensuring that project quality and schedule are not affected. Summary of the Invention

[0003] This invention provides a method for online monitoring of the slump of poured and mixed concrete, mainly comprising:

[0004] Several concrete surface moisture sensors and wind speed sensors are deployed at the concrete pouring site. The sensor sampling frequency is preset, and the measurement range covers the concrete surface and the environment to obtain concrete surface moisture evaporation rate data and wind speed data.

[0005] The collected data on the rate of water evaporation from the concrete surface and wind speed data are uploaded to the data processing server in real time and stored in a time series format to form time series data on water evaporation and wind speed. The wind speed time series data is processed into wind speed data at different scales, and the wind speed data is divided into high-frequency components and low-frequency components according to different scales.

[0006] The high-frequency and low-frequency components of the wind speed data are compared with the preset gust identification threshold and turbulence identification threshold, respectively, to determine whether gusts and turbulence appear in the wind speed data. If they do, the start and end times and duration of the gusts and turbulence are recorded. Based on the recorded start and end times and duration of the wind speed, the time distribution of the gust wind speed pattern and the turbulence wind speed pattern is obtained, forming a wind speed pattern subsequence.

[0007] The time series data of water evaporation is divided according to the time distribution of different wind speed patterns to obtain the evaporation rate subsequences under different wind speed patterns. By comparing the mean and standard deviation of the evaporation rate subsequences under different wind speed patterns, the influence of different wind speed patterns on the water evaporation rate is quantified to obtain the statistical characteristics of the evaporation rate under different wind speed patterns.

[0008] Based on the statistical features of different wind speed patterns, using the statistical features of wind speed patterns as input and the statistical features of evaporation rate under the corresponding wind speed patterns as output, a prediction model for different wind speed patterns is constructed.

[0009] After the newly collected wind speed data is added, the wind speed pattern corresponding to the newly added wind speed data is determined, and the predicted value of the water evaporation rate on the concrete surface is obtained according to the prediction model under the corresponding wind speed pattern.

[0010] A concrete slump prediction model is pre-built. The predicted value of the water evaporation rate on the concrete surface is used as input. The input water evaporation rate data is analyzed and predicted to obtain the predicted value of the concrete slump. If the predicted concrete slump value exceeds or falls below the preset qualified range, the concrete slump is determined to be unqualified and an early warning signal is issued.

[0011] After the warning signal is triggered, the concrete mix ratio adjustment plan is determined based on the value of the predicted concrete slump exceeding or falling below the preset qualified range. The concrete mix ratio parameters of the adjustment plan are input into the concrete production equipment, and the surface moisture evaporation rate data of the adjusted concrete is continuously acquired and returned to the input of the concrete slump prediction model for the next online prediction of concrete slump.

[0012] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0013] This invention discloses a method for online monitoring of the slump of poured concrete. The method involves deploying sensors at the concrete pouring site to collect real-time data on the rate of water evaporation from the concrete surface and wind speed, and using wavelet transform to identify different wind speed patterns. Based on this data, the invention constructs a correlation model between wind speed patterns and water evaporation rates, thereby predicting the concrete slump. When the predicted slump is unqualified, the invention automatically triggers an early warning and provides a concrete mix adjustment plan, achieving closed-loop control of concrete quality. This method innovatively integrates environmental factors, material properties, and the production process, achieving accurate prediction and real-time control of concrete slump, effectively improving the quality and efficiency of concrete construction, and providing a new technological path for intelligent concrete production. Attached Figure Description

[0014] Figure 1 This is a flowchart of a method for online monitoring of the slump of poured and mixed concrete according to the present invention.

[0015] Figure 2 This is a schematic diagram of a method for online monitoring of the slump of poured and mixed concrete according to the present invention.

[0016] Figure 3 This is another schematic diagram of a method for online monitoring of the slump of poured and mixed concrete according to the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] like Figure 1-3 The method for online monitoring of the slump of poured and mixed concrete in this embodiment may specifically include:

[0019] Step S101: Install several concrete surface moisture sensors and wind speed sensors at the concrete pouring site, preset the sensor sampling frequency, measure the concrete surface and environment within a preset range, and obtain concrete surface moisture evaporation rate data and wind speed data.

[0020] The concrete surface is divided into measurement sub-regions based on the sensor sampling area grid diagram. The time-series monitoring values ​​of the sensor array are obtained using the data acquisition module. The time-series monitoring values ​​are spatially interpolated using the Kriging interpolation algorithm to obtain the moisture distribution map and wind speed distribution map of the concrete surface. The difference in moisture content between adjacent sampling times in the moisture distribution map is divided by the sampling time interval to obtain the moisture evaporation rate data of the measurement point. If the wind speed sensor detection value is greater than the preset wind speed threshold value, Gaussian filtering is used to denoise the moisture evaporation rate data to obtain the corrected moisture evaporation rate data.

[0021] Specifically, the concrete surface is divided into several measurement sub-regions according to the sensor sampling area grid. A concrete surface moisture sensor and a wind speed sensor are deployed at the center point of each sub-region. A data acquisition module periodically collects data from the sensor array within the region at preset sampling time intervals to obtain time-series monitoring values. A sensor calibration module performs zero-point drift correction and linearity compensation on the collected concrete surface moisture and wind speed sensor data. A Kriging interpolation algorithm is used to spatially interpolate the data between the measurement sub-regions to obtain the moisture distribution map and wind speed distribution map of the entire concrete surface. A moving average filter is applied to the time-series monitoring values ​​to eliminate random fluctuations. The moisture evaporation rate data for each measurement point is calculated by dividing the difference in concrete surface moisture content between adjacent sampling times by the sampling time interval. If the wind speed sensor reading is greater than a preset wind speed threshold, a Gaussian filter is used to denoise the moisture evaporation rate data to obtain corrected moisture evaporation rate data. The preset wind speed threshold is determined based on the risk of rapid moisture loss from the concrete surface. A recursive least squares method was used to fit and regress the corrected moisture evaporation rate data with wind speed data, establishing a linear regression equation between moisture evaporation rate and wind speed. The independent variable was wind speed, and the dependent variable was moisture evaporation rate. Based on the established linear regression equation, the slope coefficient and intercept coefficient between the concrete surface moisture evaporation rate and wind speed were calculated, forming a quantitative parameter matrix reflecting the moisture evaporation characteristics. At the concrete construction site, sensors were deployed using a grid layout, with each measurement sub-region having a side length of 0.5 meters. For a 10m × 10m concrete surface, a total of 400 measurement sub-regions were divided. A concrete surface moisture sensor (manufactured by WIKA AG, Germany) and a three-dimensional ultrasonic anemometer (manufactured by Campbell AG, USA) were deployed at the center point of each measurement sub-region. The sensor sampling time interval was set to 1 minute. During sensor calibration, standard solutions were used to perform three-point calibration of the concrete surface moisture sensor. The standard solutions had moisture contents of 5%, 10%, and 15%, respectively, obtaining a linear relationship curve between the sensor output voltage and the moisture content. The wind speed sensor was calibrated using a standard wind tunnel with a wind speed range of 0 to 20 m / s. Calibration points were taken every 2 m / s to obtain the correspondence between the output signal and the wind speed. Spatial interpolation employed the Kriging algorithm, which considers the spatial correlation between measurement points and describes the spatial correlation characteristics using a variogram model. A spherical variogram model was chosen, with the mathematical expression γ(h) = C. O +C[1.5(h / a)-0.5(h / a) 3 ], where h is the sampling point spacing, C OThe nugget value was set to 0.01, C to the sill value to 0.95, and a to the range to 5 meters. For the collected time-series data, a 5-point moving average method was used to eliminate random fluctuations. When calculating the evaporation rate, the instantaneous evaporation rate was obtained by dividing the difference in moisture content between two adjacent sampling times by the sampling time interval of 60 seconds. When the wind speed exceeded 5 m / s, a Gaussian filter was used to denoise the evaporation rate data, with the standard deviation of the Gaussian filter set to 1.2. To establish the relationship between the evaporation rate and wind speed, a recursive least squares method was used for fitting, with the regression equation in the form y = kx + b, where y is the evaporation rate (g / m²·h), x is the wind speed (m / s), k is the slope coefficient, and b is the intercept coefficient. Typical parameter values ​​obtained through fitting experimental data were k = 2.35, b = 0.82, and the correlation coefficient R0 was [missing value]. 2 =0.94, indicating a strong linear correlation between wind speed and moisture evaporation rate. In practical applications, for a measurement sub-area, when the wind speed is 3 m / s, the moisture evaporation rate calculated according to the regression equation is 7.87 g / m²·h. When the wind speed increases to 8 m / s, the moisture evaporation rate rises to 19.62 g / m²·h. The relative error between this and the actual measured value is within ±5%, verifying the accuracy of the regression model.

[0022] Step S102: The collected concrete surface moisture evaporation rate data and wind speed data are uploaded to the data processing server in real time and stored in time series format to form moisture evaporation time series data and wind speed time series data. The wind speed time series data is processed into wind speed data at different scales, and the wind speed data is divided into high-frequency components and low-frequency components according to different scales.

[0023] A concrete surface monitoring device is used to acquire moisture evaporation rate data and wind speed data. The moisture evaporation rate data and wind speed data are timestamped according to a preset sampling period and then compressed before being transmitted to a data processing server. The compressed data is received and decompressed using a time-series database. Based on the timestamped data, missing data points are interpolated using a cubic spline interpolation algorithm to obtain complete time-series data. A median filter is used to preprocess the complete time-series data, removing measurement noise and outliers to obtain preprocessed wind speed data. Based on the preprocessed wind speed data, discrete wavelet transform decomposition is performed using the db4 wavelet basis function. The instantaneous frequency characteristics of the wind speed data are calculated using Hilbert transform. A Butterworth bandpass filter bank is used to divide the wind speed data into frequency ranges, obtaining high-frequency and low-frequency wind speed components.

[0024] Specifically, based on the concrete surface moisture evaporation rate data and wind speed data in the temporary storage area of ​​the acquisition device, timestamps are added to the data according to a preset sampling period. The timestamped data is then compressed and packaged using a lossless compression algorithm and transmitted over the network to the data processing server. A time-series database is used to decompress and restore the compressed data packets, generating time-series data for moisture evaporation rate and wind speed according to the timestamps. Missing data points are interpolated using a cubic spline interpolation algorithm to obtain a complete time series. The time-series database uses a timestamp index structure to store the data. Median filtering is applied to the complete time-series data to remove measurement noise and outliers, resulting in preprocessed wind speed time-series data. The db4 wavelet basis function is used to perform discrete wavelet transform multi-scale decomposition on the preprocessed wind speed time-series data. The signal is iteratively decomposed into four levels according to a preset decomposition level, obtaining wavelet coefficients and scale coefficients for each level. The wind speed components at each scale are reconstructed based on the obtained wavelet coefficients and scale coefficients, and the instantaneous frequency characteristics of the wind speed components at each scale are calculated using Hilbert transform. The wind speed component data was divided into frequency ranges based on instantaneous frequency characteristics. A Butterworth bandpass filter bank was used to filter the wind speed components in different frequency ranges, obtaining high-frequency and low-frequency wind speed component data, and recording the energy proportion corresponding to each frequency component. During the data acquisition process for the concrete surface moisture evaporation rate and wind speed data, a ring-shaped buffer structure was used for the temporary data storage area. The buffer size was set to 1024 data points, and the sampling period was 1 second. Data compression and transmission were triggered when the buffer data reached 512 data points. Run-length encoding in the lossless compression algorithm was used for data compression. For continuous wind speed data such as 3.5, 3.5, 3.5, 3.6, 3.6, the compressed representation was (3.5, 3)(3.6, 2), achieving a compression ratio of 2:1. The time-series database used a timestamp index tree structure to store data, with timestamp precision at the millisecond level. The leaf nodes of the index tree stored the specific data values. When data was missing, cubic spline interpolation was used for data completion, with the interpolation polynomial form being f(x) = ax. 3 +bx 2 + cx+d, where a, b, c, and d are undetermined coefficients obtained by solving a system of linear equations. For four adjacent data points (0, 1.2), (1, 1.5), (3, 2.1), and (4, 2.4), the interpolation function obtained can be used to calculate the missing value 1.8 at time x = 2. Data preprocessing uses 5-point median filtering. For the sequence data [1.2, 1.5, 5.8, 1.8, 2.1], the median value 1.8 is used to replace the original value 5.8 to eliminate abrupt interference. The db4 wavelet basis function is selected for four-level wavelet decomposition. The basis function has good time-frequency localization characteristics and a support length of 7. For wind speed data with a length of 1024 points, the first level of decomposition yields 512 low-frequency coefficients and 512 high-frequency coefficients, the second level of decomposition yields 256 low-frequency coefficients and 256 high-frequency coefficients, and so on. In frequency characteristic analysis, the Hilbert transform is used to calculate the instantaneous frequency. Let the original signal be x(t), and its Hilbert transform be y(t). Then the analytic signal z(t) = x(t) + jy(t), where j is the imaginary unit. The instantaneous phase φ(t) = arctan(y(t) / x(t)), and the instantaneous frequency f(t) = (1 / 2π)·dφ(t) / dt. For the wind speed signal, the calculated instantaneous frequency distribution range is 0-0.5Hz. Based on the instantaneous frequency characteristics, a fourth-order Butterworth bandpass filter bank is designed, dividing the 0-0.1Hz range into a low-frequency band and the 0.1-0.5Hz range into a high-frequency band. The filter frequency response function H(ω) = 1 / [1 + (ω / ωc)⁸], where ωc is the cutoff frequency and ω is the angular frequency. Energy statistics show that the low-frequency component accounts for 75% of the total energy, reflecting that wind speed changes are mainly slow and low-frequency.

[0025] Step S103: Compare the high-frequency and low-frequency components of the wind speed data with the preset gust identification threshold and turbulence identification threshold respectively to determine whether gusts and turbulence appear in the wind speed data. If they appear, record the start and end times and duration of the gusts and turbulence. Based on the recorded start and end times and duration of the wind speed, obtain the time distribution of the gust wind speed pattern and the turbulence wind speed pattern to form a wind speed pattern subsequence.

[0026] A frequency separator is used to separate the high-frequency and low-frequency components of the wind speed data. Gust event markers are obtained by comparing the high-frequency components with a threshold, and turbulence event markers are obtained by comparing the low-frequency components with a threshold. A time window processor is used to segment the wind speed data, and the event count value within each time window is obtained by statistically analyzing the gust event markers and the turbulence event markers. Feature extraction is performed on the wind speed data within the time window, and gust feature sequences and turbulence feature sequences are calculated based on the event count values. A dynamic programming algorithm is used to time-align the gust feature sequences and the turbulence feature sequences, and a wind speed pattern feature vector is generated based on the aligned feature sequences.

[0027] Specifically, an amplitude comparator is used to compare the high-frequency components of wind speed data with a preset gust identification threshold. When the high-frequency component amplitude exceeds the threshold, the gust is marked as starting; when the high-frequency component amplitude at five consecutive sampling points is below the threshold, the gust is marked as ending. The peak value and duration of the high-frequency components of wind speed within this time period are recorded. The low-frequency components of wind speed data are compared with a preset turbulence identification threshold. The root mean square (RMS) value of the low-frequency components is calculated as a cumulative turbulence intensity index. When the cumulative turbulence intensity index exceeds the preset turbulence identification threshold, the turbulence is marked as starting; when the index decreases below the threshold, the turbulence is marked as ending. The RMS value and duration of the low-frequency components within this time period are recorded. Based on the recorded gust and turbulence event time information, a fixed-length time window is set to segment the wind speed data. The number of occurrences and cumulative duration of gusts and turbulence within each time window are counted. Feature extraction is performed on the wind speed data within each time window, calculating the gust peak coefficient and turbulence intensity coefficient to generate gust feature sequences and turbulence feature sequences. A dynamic programming algorithm was used to align the gust and turbulence feature sequences over time, generating wind speed pattern subsequences based on the temporal distribution patterns of the feature sequences. Wind speed pattern feature vectors were established based on the combination relationship between gust and turbulence features in the wind speed pattern subsequences, recording the temporal distribution of various wind speed patterns. In wind speed data processing, the gust and turbulence identification thresholds were determined based on statistical analysis of extensive historical data. The gust identification threshold was set to 2.5 times the standard deviation of the wind speed. For an average wind speed of 5 m / s, with a standard deviation of 0.8 m / s, the threshold was set to 2.0 m / s. The turbulence identification threshold was based on the turbulence intensity index, with a value of 0.15, meaning that wind speed fluctuations exceeding 15% of the average wind speed were considered turbulent. In practical applications, the original wind speed data sequence [4.8, 4.9, 7.2, 7.5, 7.3, 4.7, 4.8] m / s is processed by high-frequency component extraction to obtain [0.1, 0.2, 2.4, 2.7, 2.5, -0.1, 0.0] m / s. A gust event is identified when three consecutive high-frequency component values ​​exceed 2.0 m / s, lasting for three sampling periods. Simultaneously, the low-frequency component sequence [4.7, 4.7, 4.8, 4.8, 4.8, 4.8, 4.8] m / s represents the slow variation trend of wind speed. The cumulative turbulence intensity is calculated using the root mean square (RMS) value. For the aforementioned low-frequency component sequence, the RMS value within a 10-second time window is 0.05, lower than the turbulence identification threshold of 0.15, indicating that no turbulence occurred during this period. In another data sequence, when low-frequency fluctuations such as [4.5, 4.8, 5.2, 5.5, 5.2, 4.8, 4.5] occur, the root mean square value reaches 0.18, exceeding the threshold, thus identifying a turbulence event. During feature extraction, the peak gust coefficient is defined as the ratio of the maximum instantaneous wind speed to the average wind speed. For example, if the maximum wind speed during a gust is 7.5 m / s and the average wind speed is 4.8 m / s, the peak gust coefficient is 1.56.The turbulence intensity coefficient is calculated by dividing the standard deviation of wind speed fluctuations by the average wind speed; a larger value indicates stronger turbulence. In the sample data, with a standard deviation of 0.8 m / s and an average wind speed of 5.0 m / s, the turbulence intensity coefficient is 0.16. The dynamic programming algorithm achieves time alignment of the feature sequences by constructing a state transition matrix. Taking 2 minutes of data as an example, 12 10-second time windows are set. Gust features are detected in windows 3, 4, and 5 [1.56, 1.52, 1.48], and turbulence features are detected in windows 4, 5, and 6 [0.16, 0.15, 0.14], indicating that gusts and turbulence have temporal overlap. The generated wind speed pattern feature vector includes features such as gust peak coefficient, duration, turbulence intensity coefficient, and overlap, used to characterize the time-varying characteristics of wind speed.

[0028] Step S104: Divide the water evaporation time series data into segments according to the time distribution of different wind speed modes to obtain evaporation rate subsequences under different wind speed modes. By comparing the mean and standard deviation of the evaporation rate subsequences under different wind speed modes, the influence of different wind speed modes on the water evaporation rate is quantified to obtain the statistical characteristic quantity of the evaporation rate under different wind speed modes.

[0029] The water evaporation time series data is segmented according to the timestamp alignment method. Cubic spline interpolation is used to supplement boundary data at the segmentation points, resulting in a water evaporation rate subsequence that satisfies the first derivative continuity constraint. The trapezoidal integral method is used to integrate the water evaporation rate subsequence, and the arithmetic mean of the integration result is calculated to obtain the instantaneous evaporation rate within the corresponding time period. Anomalies in the water evaporation rate subsequence are identified using a box plot method, and corrected using linear interpolation, resulting in a corrected water evaporation rate subsequence. The moment estimation method is used to calculate the central moments of the corrected water evaporation rate subsequence, and the central moments are standardized using a maximum-minimum normalization method. The standardized feature vectors are then grouped using a hierarchical clustering method to obtain statistical feature combinations of water evaporation rates under different wind speed modes.

[0030] Specifically, based on the temporal distribution data of wind speed patterns, the water evaporation time series data is segmented using timestamp alignment. At the segmentation points, cubic spline interpolation is used to supplement boundary data, with the interpolation function satisfying the first derivative continuity constraint at the boundary points, generating water evaporation rate subsequences corresponding to different wind speed patterns. The trapezoidal integral method is used to calculate the instantaneous evaporation rate of each water evaporation rate subsequence within the corresponding time period, and the arithmetic mean of the water evaporation rate subsequences is calculated based on the integration results. Outlier removal is performed on the water evaporation rate subsequences, using box plots to identify outlier data points, which are then corrected using linear interpolation to obtain the corrected water evaporation rate subsequences. The moment estimation method is used to calculate the central moments of the corrected subsequences, including the mean, variance, third central moment, and fourth central moment. The central moment data is standardized using the minimax normalization method. Based on the standardized central moment data, water evaporation rate feature vectors are constructed, and Euclidean distance is used to measure the similarity between feature vectors. A bottom-up hierarchical clustering method was used to group the feature vectors, and the inter-class and intra-class distances were calculated using the Ward minimum variance criterion. Based on the clustering results, statistical feature combinations of water evaporation rates under different wind speed modes were determined, and the mean and standard deviation of each combination were recorded. For the segmented processing of the water evaporation rate time series, cubic spline interpolation was used at boundary points to ensure data continuity. Taking the wind speed mode switching point t = 120 seconds as an example, two data points were taken before and after the switching point to form a data group.

[0031] Given the polynomial {(118,2.5),(119,2.6),(120,2.8),(121,3.2),(122,3.3)}, the interpolation function f(t) = 0.002t is obtained by solving the coefficients of the third-order polynomial. 3 -0.015t 2The function +0.12t+1.8 satisfies the continuity of the first derivative at the boundary points. In the calculation of the instantaneous evaporation rate, the trapezoidal integral method is used to process the subsequences. For the sequence data [2.5, 2.8, 3.1, 2.9, 2.7] with a time interval of 1 second, the integral values ​​between two adjacent points are (2.5+2.8)×0.5=2.65, (2.8+3.1)×0.5=2.95, and so on, to obtain the integral value of the complete sequence, thus calculating the average evaporation rate of 2.82 g / m²·s. Outlier identification uses the box plot method. The quartiles of the sequence data are calculated as Q1=2.6, Q2=2.8, Q3=3.0, and the interquartile range IQR=0.4. The outlier judgment range is set as [Q1-1.5IQR, Q3+1.5IQR], i.e., [2.0, 3.6]. The value 4.2 in the original sequence exceeds the upper limit of 3.6 and is therefore identified as an outlier. It is corrected to 3.3 using linear interpolation. The central moments reflect the statistical characteristics of the data. For the sequence [2.5, 2.8, 3.1, 2.9, 2.7], the first central moment (mean) μ = 2.8, and the second central moment (variance) σ... 2 =0.045, third-order central moment (skewness) γ = 0.12, fourth-order central moment (kurtosis) κ = 2.3. Using the maximum-minimum normalization method, x' = (x - xmin) / (xmax - xmin) maps each order of central moment to the [0,1] interval. The Euclidean distance calculation of the eigenvectors reflects the differences in evaporation characteristics under different wind speed modes. The Euclidean distance d between two eigenvectors v1 = [0.3, 0.5, 0.2, 0.4] and v2 = [0.5, 0.6, 0.3, 0.5] is √[(0.3 - 0.5)]. 2 +(0.5-0.6) 2 +(0.2-0.3) 2 +(0.4-0.5) 2 =0.283, the smaller the distance, the more similar the features. In the hierarchical clustering process, the Ward minimum variance criterion is used to merge categories. For categories A, B, and C composed of three feature vectors, the inter-class distances D(A,B) = 5.2, D(B,C) = 3.8, and D(A,C) = 6.4 are calculated. Categories B and C with the smallest distance are selected for merging. Through the iterative merging process, the final combination of water evaporation characteristics under three typical wind speed modes—stable, turbulent, and gusty—is obtained. The average evaporation rate under the stable wind speed mode is 2.8 g / m²·s, and the standard deviation is 0.2 g / m²·s.

[0032] Step S105: Based on the statistical features of different wind speed modes, using the statistical features of the wind speed modes as input and the statistical features of the evaporation rate under the corresponding wind speed modes as output, construct a prediction model for different wind speed modes.

[0033] The feature dataset is grouped according to the wind speed pattern category label. The feature dataset is then processed using the minimum-maximum standardization method to obtain standardized wind speed pattern features and evaporation rate features. The correlation between the standardized features is calculated using the Pearson correlation coefficient. If the absolute value of the correlation coefficient is greater than a preset threshold, the dimensionality-reduced wind speed pattern features and evaporation rate features are obtained. The dimensionality-reduced feature dataset is then stratified and sampled to obtain a training set and a validation set based on the proportion of the training set. Cross-validation is used to resample the dataset to obtain a training and validation data combination. A random forest prediction model is constructed based on the training and validation data combination. The optimal parameter combination is searched within a preset parameter grid space. The root mean square error and coefficient of determination are calculated for the validation set using the optimal parameter combination.

[0034] Specifically, the statistical feature dataset is grouped according to the wind speed pattern category label. Each group of feature data is normalized, and the maximum-minimum standardization method is used to map the features to the interval [0,1], resulting in standardized wind speed pattern features and evaporation rate features. The Pearson correlation coefficient is used to calculate the correlation between features, and feature combinations with absolute correlation coefficients greater than a preset threshold are selected to generate dimensionality-reduced wind speed pattern features and evaporation rate features. The dimensionality-reduced dataset is stratified and sampled, with the training set and validation set divided at a ratio of 0.8. A five-fold cross-validation method is used to resample the training set, generating multiple training and validation data combinations. A random forest prediction model is constructed, setting the number of decision trees and the maximum tree depth parameters, and using root mean square error as the optimization objective function. A traversal search is performed within the preset parameter grid space, training the prediction model for each parameter combination and recording the validation error. The parameter combination with the smallest validation error is selected as the optimal parameter. Based on the optimal parameters, a prediction model is constructed for each wind speed pattern, trained using the complete training set, and the tree structure parameters and node splitting rules of the model are recorded. The performance of the trained prediction model was evaluated using a validation set. The root mean square error and coefficient of determination between the predicted and actual values ​​were calculated, and a model validation report was generated. In the standardization of wind speed model features, the original feature data [15.2, 18.5, 12.8, 22.4, 16.7] were normalized using the maximum-minimum standardization formula x' = (x - xmin) / (xmax - xmin) to obtain standardized feature values ​​[0.25, 0.59, 0.0, 1.0, 0.41]. For the evaporation rate feature, the original data...

[0035] The values ​​[2.8, 3.2, 2.5, 3.8, 3.0] were standardized to obtain [0.23, 0.54, 0.0, 1.0, 0.38]. During feature selection, the Pearson correlation coefficient matrix was calculated, with the correlation coefficient r = (Σ(x-μx)(y-μy)) / (σxσy), where μx and μy are the means, σx and σy are the standard deviations, and x and y represent the data sequences of the two variables or features. Taking peak wind speed and evaporation rate as an example, the calculated correlation coefficient was 0.85, exceeding the preset threshold of 0.6, so this feature was retained. Conversely, the correlation coefficient between wind speed duration and evaporation rate was only 0.32, below the threshold, so this feature was removed. In dataset partitioning, stratified sampling was used to maintain the proportion of sample categories. The original dataset contains 400 stable wind speed samples, 300 turbulent wind samples, and 200 gust wind samples. After splitting the dataset by a 0.8 ratio, the training set contains 320 stable wind speed samples, 240 turbulent wind samples, and 160 gust wind samples. Five-fold cross-validation divides the training set into five equal parts, selecting four parts for training and one part for validation each time. Parameter optimization of the random forest prediction model uses a grid search method, searching for the number of decision trees within the range of [50, 100, 150, 200] and the maximum tree depth within the range of [3, 5, 7, 9], resulting in 16 parameter combinations. The model is trained for each set of parameters, and the root mean square error is calculated. Where yi is the actual value. Here, n represents the predicted value, and n is the number of samples. The optimal parameter combination was obtained through search, resulting in 150 decision trees with a maximum depth of 7. Taking a stable wind speed model as an example, the random forest model with the optimal parameters contains 150 decision trees, each with a maximum depth of 7 layers. In the first tree, the root node splits based on wind speed peak value x1 ≤ 0.5, the left child node splits based on turbulence intensity x2 ≤ 0.3, the right child node splits based on duration x3 ≤ 0.8, and so on, forming a complete tree structure. Model validation uses two metrics: root mean square error (RMSE) and coefficient of determination (COP). On the validation set, the prediction results for the stable wind speed model show RMSE = 0.15 and R² = 0.15. 2 =0.92; the prediction result of the turbulence model is RMSE=0.18, R 2 =0.89; the RMSE of the gust model prediction result is 0.21, R 2 =0.86. The predictive performance of the three models differs, which is related to the different data complexity and sample size of each model.

[0036] Step S106: After adding new wind speed data, determine the wind speed mode corresponding to the new wind speed data, and obtain the predicted value of the water evaporation rate on the concrete surface according to the prediction model under the corresponding wind speed mode.

[0037] A data acquisition device is used to perform discrete wavelet decomposition on wind speed data. Based on the wavelet decomposition, decomposition coefficients and time indices of high-frequency and low-frequency sub-signals are obtained. Statistical parameters and energy distribution characteristics are calculated based on the decomposition coefficients. The statistical parameters are processed using the maximum-minimum standardization method to obtain a standardized wind speed feature vector. Kernel similarity is calculated between the wind speed feature vector and the support vectors in a support vector machine classifier. The wind speed pattern category to which the wind speed data belongs is determined by the kernel similarity and classification weight. A corresponding prediction model is selected based on the wind speed pattern category, and the prediction model is used to calculate the predicted value of the water evaporation rate on the concrete surface using the wind speed feature vector.

[0038] Specifically, newly added wind speed data is acquired using a data acquisition device. Based on a preset sampling period, the wind speed data undergoes a four-level discrete wavelet decomposition. The db4 wavelet basis function is used to extract high-frequency and low-frequency sub-signals from the wind speed data, and the decomposition coefficients and corresponding time indices for each level are recorded. Statistical parameters, including mean, standard deviation, kurtosis, and skewness, are calculated for each wavelet decomposition coefficient. The maximum-minimum standardization method is used to normalize these parameters, generating standardized wind speed feature parameters. The cumulative energy distribution of the wind speed data is calculated based on a sliding time window, and the energy proportion of each frequency band is extracted. A feature vector is constructed by combining this with the standardized wind speed feature parameters. A pre-trained support vector machine classifier is called, and the radial basis function kernel function is used to calculate the kernel similarity between the feature vector and the support vectors of each class. Based on the kernel similarity and classification weights, the wind speed pattern category to which the newly added wind speed data belongs is determined. Based on the identified wind speed pattern category, the corresponding prediction model is selected, and the wind speed feature vector is input into the prediction model to obtain the predicted value of the water evaporation rate at a standardized scale. Inverse normalization transformation is used to restore the predicted values ​​under the standardized scale to the original numerical range, obtaining the predicted values ​​of concrete surface moisture evaporation rate in actual physical units. The confidence interval of the predicted values ​​is calculated, and the reliability of the prediction results is judged based on a preset confidence threshold. The prediction time, predicted value, and confidence interval are recorded. In wind speed data processing, four-level discrete wavelet decomposition is used to perform multi-scale analysis on the wind speed sequence with a sampling period of 1 second. Taking the measured wind speed data [5.2, 5.5, 7.8, 7.6, 5.3, 5.4] m / s as an example, the db4 wavelet basis function is used for decomposition. The first level of decomposition obtains the low-frequency coefficients.

[0039] The second-level decomposition yields lower-frequency trend features, defined by the high-frequency coefficients [5.35, 7.7, 5.35] and [0.15, -0.1, -0.05]. In the statistical parameter calculation, eigenvalues ​​are obtained from the wavelet coefficient sequence. Taking the first-level high-frequency coefficients as an example, the mean μ = 0.0, standard deviation σ = 0.13, kurtosis k = 2.8, and skewness s = 0.15. These statistical values ​​are standardized to the [0, 1] interval after being normalized to their maximum and minimum values, forming standardized feature parameters.

[0040] [0.5, 0.65, 0.72, 0.58]. The energy distribution of the wind speed signal reflects the contribution of different frequency components. A 20-second sliding time window was used to calculate the cumulative energy; low-frequency components accounted for 72% of the total energy, mid-frequency components for 21%, and high-frequency components for 7%. Combined with standardized feature parameters, a 9-dimensional feature vector was constructed.

[0041] [0.5, 0.65, 0.72, 0.58, 0.82, 0.65, 0.72, 0.21, 0.07]. Support Vector Machine (SVM) classification uses the radial basis function K(x,y) = exp(-γ||xy||2), where γ = 0.1, and ||xy||2 is the squared Euclidean distance between x and y. The kernel similarity between the feature vector and the support vectors of each class is calculated, yielding similarity values ​​[0.82, 0.35, 0.28]. The maximum similarity of 0.82 corresponds to a stable wind speed pattern, indicating that this data segment belongs to the stable wind speed type. The prediction model inputs the standardized feature vector and outputs a predicted value of 0.45 under standardized scaling. Through inverse normalization transformation y = y'(ymax - ymin) + ymin, where y' is the standardized predicted value, ymax = 5.0, and ymin = 1.0, the actual evaporation rate prediction value of 2.8 g / m²·s is calculated. The confidence interval calculation is based on the statistical distribution of the prediction error. At a 95% confidence level, the interval estimate of the predicted value is [2.5, 3.1] g / m²·s. This interval width is less than the preset threshold of 1.0 g / m²·s, indicating that the prediction result has good reliability. The complete prediction record includes the prediction time 2024-01-15 10:30:00, the predicted value 2.8, the upper confidence limit 3.1, and the lower confidence limit 2.5, all in g / m²·s. Through this multi-level data analysis and processing, a complete transformation from raw wind speed data to moisture evaporation rate prediction is achieved. The prediction result includes both point estimates and interval estimates, providing more comprehensive prediction information.

[0042] Step S107: A concrete slump prediction model is pre-constructed. The predicted value of the water evaporation rate on the concrete surface is used as input. The input water evaporation rate data is analyzed and predicted to obtain the predicted value of the concrete slump. If the predicted concrete slump value exceeds or falls below the preset qualified range, the concrete slump is determined to be unqualified, and an early warning signal is issued.

[0043] The cumulative moving average method is used to smooth the predicted value sequence of water evaporation rate to obtain smoothed evaporation rate change rate data. Based on the evaporation rate change rate data, an exponentially weighted fusion is performed to construct a time-series feature vector reflecting the evolution characteristics of the evaporation rate. This time-series feature vector is input into a pre-trained long short-term memory network, and the slump prediction value is obtained by outputting the state through the network's memory units. If the confidence interval width of the slump prediction value exceeds a preset threshold, a local linear regression method is used to correct the prediction value. The corrected prediction value is compared with a preset acceptable range to obtain a warning marker of the corresponding level.

[0044] Specifically, the cumulative moving average method is used to smooth the predicted evaporation rate sequence. Based on a preset sampling period, the smoothed evaporation rate time series data is calculated, and a difference algorithm is used to extract the first and second rates of change of the evaporation rate. The cumulative change in the evaporation rate is calculated, and the rate of change data is fused using an exponential weighting method to construct a time-series feature vector reflecting the evolution of the evaporation rate. This time-series feature vector is input into a pre-trained Long Short-Term Memory (LSTM) network, which includes an input layer, hidden layers, and an output layer. The hidden layers consist of memory units and forget gates, recording the temporal variation of the concrete slump. The predicted slump value is calculated based on the output state of the memory units, and the reliability of the prediction results is evaluated using a preset confidence interval method. When the prediction interval width exceeds a threshold, it is marked as a low-confidence prediction result. For low-confidence prediction results, a local linear regression method is used to correct the prediction value, and the corrected slump prediction value is calculated based on historical data within the neighborhood of the prediction point. The corrected slump prediction value is compared with the upper and lower limits of a preset acceptable range. When the predicted value deviates from the acceptable range, a corresponding warning label is generated. The warning level is determined based on the severity of the warning markers; a slight deviation corresponds to a yellow warning, and a severe deviation corresponds to a red warning. The warning information is transmitted to the monitoring terminal via a data communication interface. In the processing of moisture evaporation rate data, the original sequence [2.8, 3.2, 3.5, 3.3, 2.9, 2.7] g / m²·s is smoothed using a 5-point moving average to obtain the smoothed sequence [3.14, 3.12]. The first-order difference [-0.02] reflects the trend of change, and the second-order difference reflects the acceleration of change. An exponential weight α = 0.7 is used to weight the rate of change, constructing a three-dimensional vector [3.12, -0.02, 0] reflecting the dynamic characteristics of evaporation. The core of the Long Short-Term Memory (LSTM) network lies in the state control of the memory units. Taking the prediction of collapse as an example, the network input layer receives feature vectors, and the hidden layer contains 32 memory units. Each memory unit controls the flow of information through an input gate, a forget gate, and an output gate. The input gate determines the proportion of new information received, the forget gate controls the degree of retention of historical information, and the output gate adjusts the amount of information output. When new evaporation rate data is input, the memory unit weighs the importance of the new and old information. In slump prediction, the confidence interval uses three times the mean square error as the interval radius. For a predicted value of 180 mm and a prediction standard deviation of 5 mm, the confidence interval is [165, 195] mm. When the interval width exceeds a preset threshold of 40 mm, it indicates that the prediction result has low reliability and needs to be corrected. The correction uses local linear regression, selecting five historical data points before and after the prediction point, fitting a local linear equation y = kx + b, where k is the slope and b is the intercept, and replacing the original predicted value with the fitted value. The early warning mechanism is based on the acceptable range of concrete slump.For the preset acceptable range [160, 200] mm, a yellow warning is triggered when the predicted value is within the range of [140, 160) or (200, 220], and a red warning is triggered when the predicted value exceeds the range of [140, 220]. Taking a real-world case as an example, the corrected slump prediction value is 145 mm, falling within the yellow warning range. The system generates a warning message containing the timestamp 2024-01-15 14:30:00, the predicted value 145, and the warning level yellow. The warning message is transmitted to the monitoring terminal in JSON format via the data communication interface, containing the necessary warning elements: {"timestamp":"2024-01-15 14:30:00","s `lump_predict":145,"confidence_interval":[140,150],"alert_level":"yellow"}`. After receiving the early warning information, the monitoring terminal displays corresponding visual cues according to the warning level. A yellow warning uses a flashing yellow indicator, while a red warning uses a flashing red indicator accompanied by an audible alarm. Through this multi-layered data analysis and early warning mechanism, predictions from evaporation rate to slump are achieved, and a complete early warning response process is established, providing timely early warning information for concrete quality control. The hierarchical setting and display method of the early warning mechanism intuitively reflect the severity of quality risks.

[0045] Step S108: After the warning signal is triggered, the concrete mix ratio adjustment plan is determined based on the value of the predicted concrete slump exceeding or falling below the preset qualified range. The concrete mix ratio parameters of the adjustment plan are input into the concrete production equipment, and the surface moisture evaporation rate data of the adjusted concrete is continuously acquired and returned to the input of the concrete slump prediction model for the next online prediction of concrete slump.

[0046] The deviation between the predicted concrete slump value and the preset acceptable range is fuzzified using a triangular membership function. Based on the deviation value and preset fuzzy rules, the water-cement ratio adjustment is obtained. Based on the water-cement ratio adjustment, the adjustment values ​​for cement dosage, aggregate dosage, and admixture dosage are obtained using a weighted least squares regression equation for mix proportion parameters. These adjustment values ​​are converted into equipment control commands, encapsulated via an industrial Ethernet protocol, and transmitted to the concrete production equipment. A data acquisition device is used to acquire the adjusted surface moisture data of the concrete production equipment. A moisture evaporation rate data sequence is calculated based on a preset sampling period. This moisture evaporation rate data sequence is then digitally filtered and returned to the slump predictor.

[0047] Specifically, based on the deviation between the predicted concrete slump value and the preset acceptable range, a triangular membership function is used to fuzzify the deviation value. The water-cement ratio adjustment is calculated based on preset fuzzy rules, which include the correspondence between the degree of deviation and the adjustment magnitude. A weighted least squares regression equation for the concrete mix proportions is constructed, using the water-cement ratio adjustment as the independent variable. Adjustments for cement content, coarse and fine aggregate content, and admixture dosage are calculated to generate a mix proportion parameter combination that meets workability constraints. This mix proportion parameter combination is converted into equipment control commands, encapsulated using the industrial Ethernet protocol, and transmitted to the concrete production equipment via a serial communication interface. The execution status information returned by the production equipment is received, and the execution result of the mix proportion adjustment command is determined based on the status code. The actual written value of the adjustment parameters is recorded. A data acquisition device is used to acquire the surface moisture data of the adjusted concrete, and the moisture evaporation rate per unit time is calculated according to a preset sampling period. The moisture evaporation rate data is digitally filtered to eliminate the influence of random fluctuations, resulting in a stable evaporation rate data sequence. A stable evaporation rate data sequence is returned to the slump predictor, and the predicted data is updated using a sliding time window method to achieve dynamic updates of the prediction results. The deviation between the predicted and measured values ​​is recorded. When the prediction deviation exceeds the threshold three times consecutively, online correction of the predictor parameters is triggered, updating the predictor's internal parameters. When the predicted concrete slump value deviates from the preset acceptable range, a triangular membership function is used for fuzzy processing. Taking a measured deviation of -25 mm as an example, the membership function interval is set to [-40, 0], with a vertex value of -20, and the deviation membership value is calculated to be 0.75. According to the fuzzy rule "if the deviation is negatively large, then the water-cement ratio adjustment is positively moderate," the water-cement ratio adjustment is mapped to 0.02. When constructing the regression equation for the mix proportion parameters using the weighted least squares method, the water-cement ratio x is used as the independent variable, the mix proportion parameter y is used as the dependent variable, the weight w reflects the importance of the parameter, xi is the water-cement ratio at the i-th observation point, and yi is the mix proportion parameter at the i-th observation point. For the equation y = ax + b, minimize the weighted sum of squared errors Σw(yi - axi - b). 2Solve for coefficients a and b. In the example, the regression equation for cement usage is y = -150x + 550. Substituting the water-cement ratio adjustment of 0.02, the calculated cement usage reduction is 3 kg per cubic meter. The mix proportion parameter adjustment command is transmitted using the Industrial Ethernet Modbus TCP protocol. The command frame contains function code 0x10 (write multiple registers), starting address 0x1000, number of registers 0x0004, data length 0x08, and mix proportion parameter value. After execution, the device returns a response frame. Status code 0x00 indicates successful execution, and 0x04 indicates an execution error. The data acquisition device collects surface moisture data every 30 seconds, continuously collecting 5 sets of data [12.5%, 12.2%, 11.8%, 11.5%, 11.2%]. The moisture evaporation rate is calculated to be 2.6 g / m²·min using time difference. A 5-point median filter is used to eliminate abnormal fluctuations. The filtered stable sequence is more suitable for prediction calculations. In the predictor parameter calibration, the deviation sequence [-8, -10, -12] mm between three consecutive predicted values ​​and measured values ​​was recorded. The root mean square error (RMSE) of 10.2 mm exceeded the preset threshold of 8 mm. After triggering calibration, the predictor's weight parameters were updated to make the prediction results closer to the actual trend. The closed-loop control process demonstrated a complete chain from slump prediction and mix proportion adjustment to effect verification. Taking an adjustment as an example, the predicted slump value of 145 mm was lower than the acceptable lower limit of 160 mm. The mix proportion adjustment was calculated through fuzzy control, and after the adjustment was executed, the moisture content was continuously monitored. The new predicted value of 162 mm was within the acceptable range, verifying the effectiveness of the adjustment. The entire process achieved automated control of prediction-adjustment-verification, ensuring that the concrete performance continuously met the requirements. In practical applications, this control scheme, through precise mathematical models and reliable data transmission, achieved real-time monitoring and automatic adjustment of concrete performance, significantly improving the stability of the production process and the consistency of product quality. Through the closed-loop feedback mechanism, deviations can be detected and corrected in a timely manner, avoiding losses caused by quality fluctuations.

[0048] It should be noted that the above examples are merely some specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments and many variations are possible. All variations that can be directly derived or conceived by those skilled in the art from the content disclosed in this invention should be considered within the scope of protection of this invention.

Claims

1. A method for online monitoring of slump of casted concrete, characterized in that, The method includes: deploying several concrete surface moisture sensors and wind speed sensors at the concrete pouring site; presetting the sensor sampling frequency; measuring the concrete surface and environment within a preset range to obtain concrete surface moisture evaporation rate data and wind speed data; uploading the collected concrete surface moisture evaporation rate data and wind speed data to a data processing server in real time and storing them in a time series format to form moisture evaporation time series data and wind speed time series data; processing the wind speed time series data into wind speed data at different scales; and dividing the wind speed data into high-frequency and low-frequency components according to different scales; and processing the wind speed data... High-frequency and low-frequency components are compared with preset gust and turbulence identification thresholds, respectively, to determine whether gusts and turbulence appear in the wind speed data. If they do, the start and end times and durations of the gusts and turbulence are recorded. Based on the recorded start and end times and durations, the temporal distribution of gust and turbulence wind speed patterns is obtained, forming wind speed pattern subsequences. The water evaporation time series data is segmented according to the temporal distribution of different wind speed patterns to obtain evaporation rate subsequences corresponding to different wind speed patterns. By comparing the mean and standard deviation of the evaporation rate subsequences under different wind speed patterns, the evaporation rate is quantified. The influence of different wind speed patterns on the water evaporation rate was investigated to obtain statistical characteristics of the evaporation rate under different wind speed patterns. Based on these statistical characteristics, a prediction model for different wind speed patterns was constructed, using the statistical characteristics of the wind speed patterns as input and the statistical characteristics of the evaporation rate under the corresponding wind speed patterns as output. After newly collected wind speed data, the wind speed pattern corresponding to the new wind speed data was determined, and the predicted value of the water evaporation rate on the concrete surface was obtained according to the prediction model under the corresponding wind speed pattern. A concrete slump prediction model was pre-constructed, using the predicted value of the water evaporation rate on the concrete surface as input. The system analyzes and predicts the water evaporation rate data to obtain the predicted value of concrete slump. If the predicted concrete slump value exceeds or falls below the preset acceptable range, the concrete slump is deemed unacceptable, and an early warning signal is issued. After the early warning signal is triggered, an adjustment plan for the concrete mix ratio is determined based on the value of the predicted concrete slump exceeding or falling below the preset acceptable range. The concrete mix ratio parameters of the adjustment plan are input into the concrete production equipment, and the surface water evaporation rate data of the adjusted concrete is continuously acquired and returned to the input of the concrete slump prediction model for the next online prediction of concrete slump.

2. The method of claim 1, wherein, The process of deploying several concrete surface moisture sensors and wind speed sensors at the concrete pouring site, setting a preset sensor sampling frequency, and measuring within a preset range covering the concrete surface and environment to obtain concrete surface moisture evaporation rate data and wind speed data includes: dividing the concrete surface into measurement sub-regions according to a sensor sampling area grid division map; acquiring time-series monitoring values ​​of the sensor array using a data acquisition module; performing spatial interpolation calculations on the time-series monitoring values ​​using a Kriging interpolation algorithm to obtain a moisture distribution map and a wind speed distribution map of the concrete surface; dividing the difference in moisture content between adjacent sampling times in the moisture distribution map by the sampling time interval to obtain the moisture evaporation rate data at the measurement point; and if the wind speed sensor detection value is greater than a preset wind speed threshold value, using Gaussian filtering to denoise the moisture evaporation rate data to obtain corrected moisture evaporation rate data.

3. The method of claim 1, wherein, The process involves uploading the collected concrete surface moisture evaporation rate data and wind speed data to a data processing server in real time and storing them in a time-series format to form moisture evaporation time-series data and wind speed time-series data. The wind speed time-series data is then processed into wind speed data at different scales, and based on these scales, the wind speed data is divided into high-frequency and low-frequency components. This includes: acquiring moisture evaporation rate data and wind speed data using a concrete surface monitoring device; adding timestamps to the moisture evaporation rate data and wind speed data according to a preset sampling period before compressing and transmitting them to the data processing server; receiving the compressed data and decompressing it using a time-series database; and using a cubic spline interpolation algorithm to interpolate missing data points based on the timestamps to obtain complete time-series data. The complete time series data is preprocessed using a median filter to remove measurement noise and outliers, resulting in preprocessed wind speed data. Based on the preprocessed wind speed data, discrete wavelet transform decomposition is performed using the db4 wavelet basis function. The instantaneous frequency characteristics of the wind speed data are calculated using Hilbert transform. Finally, a Butterworth bandpass filter bank is used to divide the wind speed data into frequency ranges, obtaining high-frequency and low-frequency wind speed components.

4. The method according to claim 1, characterized in that, The process involves comparing the high-frequency and low-frequency components of the wind speed data with preset gust and turbulence identification thresholds, respectively, to determine whether gusts and turbulence occur in the wind speed data. If they do, the start and end times and durations of the gusts and turbulence are recorded. Based on the recorded start and end times and durations of the wind speed occurrences, the temporal distribution of the gust and turbulence wind speed patterns is obtained, forming a wind speed pattern subsequence. This includes: using a frequency separator to separate the high-frequency and low-frequency components of the wind speed data; obtaining gust event markers based on the comparison of the high-frequency components with thresholds; obtaining turbulence event markers based on the comparison of the low-frequency components with thresholds; and using a time window processor to segment the wind speed data, and statistically obtaining the event count value within each time window based on the gust and turbulence event markers. Feature extraction is performed on the wind speed data within the time window, and gust feature sequences and turbulence feature sequences are calculated based on the event count values. A dynamic programming algorithm is used to time-align the gust feature sequences and turbulence feature sequences, and a wind speed pattern feature vector is generated based on the aligned feature sequences.

5. The method according to claim 1, characterized in that, The process involves segmenting the water evaporation time series data according to the time distribution of different wind speed patterns to obtain evaporation rate subsequences corresponding to different wind speed patterns. By comparing the mean and standard deviation of the evaporation rate subsequences under different wind speed patterns, the influence of different wind speed patterns on the water evaporation rate is quantified to obtain statistical characteristics of the evaporation rate under different wind speed patterns. This includes: segmenting the water evaporation time series data according to the timestamp alignment method; using cubic spline interpolation to supplement boundary data at the segmentation points to obtain water evaporation rate subsequences that satisfy the first derivative continuity constraint; and using the trapezoidal integral method to analyze the water evaporation rate subsequences. The sequence is integrated, and the arithmetic mean of the water evaporation rate subsequence is calculated from the integration result to obtain the instantaneous evaporation rate within the corresponding time period. Outliers in the water evaporation rate subsequence are identified using a box plot method, and linear interpolation is used to correct these outliers, resulting in a corrected water evaporation rate subsequence. The central moments of the corrected water evaporation rate subsequence are calculated using a moment estimation method, and then standardized using a maximum-minimum normalization method. Finally, the standardized feature vectors are grouped using a hierarchical clustering method to obtain statistical feature combinations of water evaporation rates under different wind speed modes.

6. The method according to claim 1, characterized in that, The method, based on statistical features of different wind speed patterns, uses statistical features of wind speed patterns as input and statistical features of evaporation rate under the corresponding wind speed patterns as output to construct prediction models for different wind speed patterns. This includes: grouping the feature dataset according to wind speed pattern category labels; processing the feature dataset using the minimax normalization method to obtain standardized wind speed pattern features and evaporation rate features; calculating the correlation between the standardized features using the Pearson correlation coefficient; if the absolute value of the correlation coefficient is greater than a preset threshold, obtaining the dimensionality-reduced wind speed pattern features and evaporation rate features; performing stratified sampling on the dimensionality-reduced feature dataset, dividing it into training and validation sets based on the proportion of the training set; resampling the training set using cross-validation to obtain a training and validation data combination; constructing a random forest prediction model based on the training and validation data combination; traversing and searching for the optimal parameter combination within a preset parameter grid space; and calculating the root mean square error and coefficient of determination for the validation set using the optimal parameter combination.

7. The method according to claim 1, characterized in that, After acquiring the newly added wind speed data, the wind speed pattern corresponding to the newly acquired wind speed data is determined, and the predicted value of the water evaporation rate on the concrete surface is obtained according to the prediction model under the corresponding wind speed pattern. This includes: using a data acquisition device to perform discrete wavelet decomposition on the wind speed data, obtaining the decomposition coefficient values ​​and time index of the high-frequency sub-signals and low-frequency sub-signals based on the wavelet decomposition; calculating statistical parameters and energy distribution characteristics based on the decomposition coefficient values, and processing the statistical parameters using the maximum-minimum-maximum standardization method to obtain the standardized wind speed feature vector. The kernel similarity is calculated between the wind speed feature vector and the support vector in the support vector machine classifier. The wind speed pattern category to which the wind speed data belongs is determined by the kernel similarity and the classification weight. The corresponding prediction model is selected according to the wind speed pattern category, and the prediction model is used to calculate the wind speed feature vector to obtain the predicted value of the water evaporation rate on the concrete surface.

8. The method according to claim 1, characterized in that, The pre-built concrete slump prediction model takes the predicted value of the concrete surface moisture evaporation rate as input, analyzes and predicts the input moisture evaporation rate data, and obtains the predicted value of the concrete slump. If the predicted concrete slump value exceeds or falls below a preset acceptable range, the concrete slump is determined to be unacceptable, and an early warning signal is issued. This includes: smoothing the moisture evaporation rate prediction value sequence using a cumulative moving average method to obtain smoothed evaporation rate change rate data; performing exponential weighted fusion based on the evaporation rate change rate data to construct a time-series feature vector reflecting the evolution characteristics of the evaporation rate; inputting the time-series feature vector into a pre-trained long short-term memory network, and obtaining the slump prediction value through the output state of the network's memory units; if the confidence interval width of the slump prediction value exceeds a preset threshold, correcting the prediction value using a local linear regression method, and comparing the corrected prediction value with the preset acceptable range to obtain a corresponding level of early warning label.

9. The method according to claim 1, characterized in that, After the warning signal is triggered, an adjustment plan for the concrete mix proportion is determined based on whether the predicted concrete slump value exceeds or falls below the preset acceptable range. The concrete mix proportion parameters of the adjustment plan are input into the concrete production equipment, and the surface moisture evaporation rate data of the adjusted concrete is continuously acquired and returned to the input of the concrete slump prediction model for the next online prediction of concrete slump. This includes: using a triangular membership function to fuzzify the deviation between the predicted concrete slump value and the preset acceptable range; obtaining the water-cement ratio adjustment amount based on the deviation value and preset fuzzy rules; obtaining the adjustment values ​​of cement dosage, aggregate dosage, and admixture dosage based on the water-cement ratio adjustment amount using a weighted least squares regression equation for the mix proportion parameters; converting the adjustment values ​​into equipment control commands, encapsulating them through an industrial Ethernet protocol, and transmitting them to the concrete production equipment; acquiring the adjusted surface moisture data of the concrete production equipment using a data acquisition device, calculating the moisture evaporation rate data sequence according to a preset sampling period, and returning the moisture evaporation rate data sequence to the slump predictor after digital filtering.

Images